commit 7bbeb2d1442af00b3dc658af30c442d83cd767c0
parent def5d9802f375b1ba9da443d8a889c175e2bdd3f
Author: Martin Herkt <lachs0r@srsfckn.biz>
Date: Fri, 27 Oct 2017 05:22:11 +0200
Add NSFW detection
Diffstat:
7 files changed, 3666 insertions(+), 10 deletions(-)
diff --git a/README.rst b/README.rst
@@ -25,6 +25,17 @@ now and then.
Before running the service for the first time, run ``./fhost.py db upgrade``.
+NSFW Detection
+--------------
+
+0x0 supports classification of NSFW content via Yahoo’s open_nsfw Caffe
+neural network model. This works for images and video files and requires
+the following:
+
+* Caffe Python module (built for Python 3)
+* ``ffmpegthumbnailer`` executable in ``$PATH``
+
+
FAQ
---
diff --git a/fhost.py b/fhost.py
@@ -44,6 +44,13 @@ app.config["FHOST_MIME_BLACKLIST"] = [
app.config["FHOST_UPLOAD_BLACKLIST"] = "tornodes.txt"
+app.config["NSFW_DETECT"] = True
+app.config["NSFW_THRESHOLD"] = 0.7
+
+if app.config["NSFW_DETECT"]:
+ from nsfw_detect import NSFWDetector
+ nsfw = NSFWDetector()
+
try:
mimedetect = Magic(mime=True, mime_encoding=False)
except:
@@ -72,6 +79,9 @@ class URL(db.Model):
def getname(self):
return su.enbase(self.id, 1)
+ def geturl(self):
+ return url_for("get", path=self.getname(), _external=True) + "\n"
+
class File(db.Model):
id = db.Column(db.Integer, primary_key = True)
sha256 = db.Column(db.String, unique = True)
@@ -79,23 +89,29 @@ class File(db.Model):
mime = db.Column(db.UnicodeText)
addr = db.Column(db.UnicodeText)
removed = db.Column(db.Boolean, default=False)
+ nsfw_score = db.Column(db.Float)
- def __init__(self, sha256, ext, mime, addr):
+ def __init__(self, sha256, ext, mime, addr, nsfw_score):
self.sha256 = sha256
self.ext = ext
self.mime = mime
self.addr = addr
+ self.nsfw_score = nsfw_score
def getname(self):
return u"{0}{1}".format(su.enbase(self.id, 1), self.ext)
+ def geturl(self):
+ n = self.getname()
+
+ if self.nsfw_score and self.nsfw_score > app.config["NSFW_THRESHOLD"]:
+ return url_for("get", path=n, _external=True, _anchor="nsfw") + "\n"
+ else:
+ return url_for("get", path=n, _external=True) + "\n"
def getpath(fn):
return os.path.join(app.config["FHOST_STORAGE_PATH"], fn)
-def geturl(p):
- return url_for("get", path=p, _external=True) + "\n"
-
def fhost_url(scheme=None):
if not scheme:
return url_for(".fhost", _external=True).rstrip("/")
@@ -115,13 +131,13 @@ def shorten(url):
existing = URL.query.filter_by(url=url).first()
if existing:
- return geturl(existing.getname())
+ return existing.geturl()
else:
u = URL(url)
db.session.add(u)
db.session.commit()
- return geturl(u.getname())
+ return u.geturl()
def in_upload_bl(addr):
if os.path.isfile(app.config["FHOST_UPLOAD_BLACKLIST"]):
@@ -152,11 +168,15 @@ def store_file(f, addr):
with open(epath, "wb") as of:
of.write(data)
+ if existing.nsfw_score == None:
+ if app.config["NSFW_DETECT"]:
+ existing.nsfw_score = nsfw.detect(epath)
+
os.utime(epath, None)
existing.addr = addr
db.session.commit()
- return geturl(existing.getname())
+ return existing.geturl()
else:
guessmime = mimedetect.from_buffer(data)
@@ -186,14 +206,21 @@ def store_file(f, addr):
if not ext:
ext = ".bin"
- with open(getpath(digest), "wb") as of:
+ spath = getpath(digest)
+
+ with open(spath, "wb") as of:
of.write(data)
- sf = File(digest, ext, mime, addr)
+ if app.config["NSFW_DETECT"]:
+ nsfw_score = nsfw.detect(spath)
+ else:
+ nsfw_score = None
+
+ sf = File(digest, ext, mime, addr, nsfw_score)
db.session.add(sf)
db.session.commit()
- return geturl(sf.getname())
+ return sf.geturl()
def store_url(url, addr):
if is_fhost_url(url):
@@ -438,6 +465,37 @@ def queryaddr(a):
for f in res:
query(su.enbase(f.id, 1))
+def nsfw_detect(f):
+ try:
+ open(f["path"], 'r').close()
+ f["nsfw_score"] = nsfw.detect(f["path"])
+ return f
+ except:
+ return None
+
+@manager.command
+def update_nsfw():
+ if not app.config["NSFW_DETECT"]:
+ print("NSFW detection is disabled in app config")
+ return 1
+
+ from multiprocessing import Pool
+ import tqdm
+
+ res = File.query.filter_by(nsfw_score=None, removed=False)
+
+ with Pool() as p:
+ results = []
+ work = [{ "path" : getpath(f.sha256), "id" : f.id} for f in res]
+
+ for r in tqdm.tqdm(p.imap_unordered(nsfw_detect, work), total=len(work)):
+ if r:
+ results.append({"id": r["id"], "nsfw_score" : r["nsfw_score"]})
+
+ db.session.bulk_update_mappings(File, results)
+ db.session.commit()
+
+
@manager.command
def querybl():
if os.path.isfile(app.config["FHOST_UPLOAD_BLACKLIST"]):
diff --git a/migrations/versions/7e246705da6a_.py b/migrations/versions/7e246705da6a_.py
@@ -0,0 +1,26 @@
+"""add NSFW score
+
+Revision ID: 7e246705da6a
+Revises: 0cd36ecdd937
+Create Date: 2017-10-27 03:07:48.179290
+
+"""
+
+# revision identifiers, used by Alembic.
+revision = '7e246705da6a'
+down_revision = '0cd36ecdd937'
+
+from alembic import op
+import sqlalchemy as sa
+
+
+def upgrade():
+ # ### commands auto generated by Alembic - please adjust! ###
+ op.add_column('file', sa.Column('nsfw_score', sa.Float(), nullable=True))
+ # ### end Alembic commands ###
+
+
+def downgrade():
+ # ### commands auto generated by Alembic - please adjust! ###
+ op.drop_column('file', 'nsfw_score')
+ # ### end Alembic commands ###
diff --git a/nsfw_detect.py b/nsfw_detect.py
@@ -0,0 +1,62 @@
+#!/usr/bin/env python3
+
+import numpy as np
+import os
+import sys
+from io import BytesIO
+from subprocess import run, PIPE, DEVNULL
+
+os.environ["GLOG_minloglevel"] = "2" # seriously :|
+import caffe
+
+class NSFWDetector:
+ def __init__(self):
+
+ npath = os.path.join(os.path.dirname(__file__), "nsfw_model")
+ self.nsfw_net = caffe.Net(os.path.join(npath, "deploy.prototxt"),
+ os.path.join(npath, "resnet_50_1by2_nsfw.caffemodel"),
+ caffe.TEST)
+ self.caffe_transformer = caffe.io.Transformer({'data': self.nsfw_net.blobs['data'].data.shape})
+ self.caffe_transformer.set_transpose('data', (2, 0, 1)) # move image channels to outermost
+ self.caffe_transformer.set_mean('data', np.array([104, 117, 123])) # subtract the dataset-mean value in each channel
+ self.caffe_transformer.set_raw_scale('data', 255) # rescale from [0, 1] to [0, 255]
+ self.caffe_transformer.set_channel_swap('data', (2, 1, 0)) # swap channels from RGB to BGR
+
+ def _compute(self, img):
+ image = caffe.io.load_image(BytesIO(img))
+
+ H, W, _ = image.shape
+ _, _, h, w = self.nsfw_net.blobs["data"].data.shape
+ h_off = int(max((H - h) / 2, 0))
+ w_off = int(max((W - w) / 2, 0))
+ crop = image[h_off:h_off + h, w_off:w_off + w, :]
+
+ transformed_image = self.caffe_transformer.preprocess('data', crop)
+ transformed_image.shape = (1,) + transformed_image.shape
+
+ input_name = self.nsfw_net.inputs[0]
+ output_layers = ["prob"]
+ all_outputs = self.nsfw_net.forward_all(blobs=output_layers,
+ **{input_name: transformed_image})
+
+ outputs = all_outputs[output_layers[0]][0].astype(float)
+
+ return outputs
+
+ def detect(self, fpath):
+ try:
+ ff = run(["ffmpegthumbnailer", "-m", "-o-", "-s256", "-t50%", "-a", "-cpng", "-i", fpath], stdout=PIPE, stderr=DEVNULL, check=True)
+ image_data = ff.stdout
+ except:
+ return -1.0
+
+ scores = self._compute(image_data)
+
+ return scores[1]
+
+if __name__ == "__main__":
+ n = NSFWDetector()
+
+ for inf in sys.argv[1:]:
+ score = n.detect(inf)
+ print(inf, score)
diff --git a/nsfw_model/LICENSE.md b/nsfw_model/LICENSE.md
@@ -0,0 +1,11 @@
+
+Copyright 2016, Yahoo Inc.
+
+Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
+
+1. Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
+
+2. Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
diff --git a/nsfw_model/deploy.prototxt b/nsfw_model/deploy.prototxt
@@ -0,0 +1,3488 @@
+name: "ResNet_50_1by2_nsfw"
+layer {
+ name: "data"
+ type: "Input"
+ top: "data"
+ input_param { shape: { dim: 1 dim: 3 dim: 224 dim: 224 } }
+}
+layer {
+ name: "conv_1"
+ type: "Convolution"
+ bottom: "data"
+ top: "conv_1"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 64
+ pad: 3
+ kernel_size: 7
+ stride: 2
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_1"
+ type: "BatchNorm"
+ bottom: "conv_1"
+ top: "conv_1"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_1"
+ type: "Scale"
+ bottom: "conv_1"
+ top: "conv_1"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "relu_1"
+ type: "ReLU"
+ bottom: "conv_1"
+ top: "conv_1"
+}
+layer {
+ name: "pool1"
+ type: "Pooling"
+ bottom: "conv_1"
+ top: "pool1"
+ pooling_param {
+ pool: MAX
+ kernel_size: 3
+ stride: 2
+ }
+}
+layer {
+ name: "conv_stage0_block0_proj_shortcut"
+ type: "Convolution"
+ bottom: "pool1"
+ top: "conv_stage0_block0_proj_shortcut"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 128
+ pad: 0
+ kernel_size: 1
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage0_block0_proj_shortcut"
+ type: "BatchNorm"
+ bottom: "conv_stage0_block0_proj_shortcut"
+ top: "conv_stage0_block0_proj_shortcut"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage0_block0_proj_shortcut"
+ type: "Scale"
+ bottom: "conv_stage0_block0_proj_shortcut"
+ top: "conv_stage0_block0_proj_shortcut"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "conv_stage0_block0_branch2a"
+ type: "Convolution"
+ bottom: "pool1"
+ top: "conv_stage0_block0_branch2a"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 32
+ pad: 0
+ kernel_size: 1
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage0_block0_branch2a"
+ type: "BatchNorm"
+ bottom: "conv_stage0_block0_branch2a"
+ top: "conv_stage0_block0_branch2a"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage0_block0_branch2a"
+ type: "Scale"
+ bottom: "conv_stage0_block0_branch2a"
+ top: "conv_stage0_block0_branch2a"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "relu_stage0_block0_branch2a"
+ type: "ReLU"
+ bottom: "conv_stage0_block0_branch2a"
+ top: "conv_stage0_block0_branch2a"
+}
+layer {
+ name: "conv_stage0_block0_branch2b"
+ type: "Convolution"
+ bottom: "conv_stage0_block0_branch2a"
+ top: "conv_stage0_block0_branch2b"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 32
+ pad: 1
+ kernel_size: 3
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage0_block0_branch2b"
+ type: "BatchNorm"
+ bottom: "conv_stage0_block0_branch2b"
+ top: "conv_stage0_block0_branch2b"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage0_block0_branch2b"
+ type: "Scale"
+ bottom: "conv_stage0_block0_branch2b"
+ top: "conv_stage0_block0_branch2b"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "relu_stage0_block0_branch2b"
+ type: "ReLU"
+ bottom: "conv_stage0_block0_branch2b"
+ top: "conv_stage0_block0_branch2b"
+}
+layer {
+ name: "conv_stage0_block0_branch2c"
+ type: "Convolution"
+ bottom: "conv_stage0_block0_branch2b"
+ top: "conv_stage0_block0_branch2c"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 128
+ pad: 0
+ kernel_size: 1
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage0_block0_branch2c"
+ type: "BatchNorm"
+ bottom: "conv_stage0_block0_branch2c"
+ top: "conv_stage0_block0_branch2c"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage0_block0_branch2c"
+ type: "Scale"
+ bottom: "conv_stage0_block0_branch2c"
+ top: "conv_stage0_block0_branch2c"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "eltwise_stage0_block0"
+ type: "Eltwise"
+ bottom: "conv_stage0_block0_proj_shortcut"
+ bottom: "conv_stage0_block0_branch2c"
+ top: "eltwise_stage0_block0"
+}
+layer {
+ name: "relu_stage0_block0"
+ type: "ReLU"
+ bottom: "eltwise_stage0_block0"
+ top: "eltwise_stage0_block0"
+}
+layer {
+ name: "conv_stage0_block1_branch2a"
+ type: "Convolution"
+ bottom: "eltwise_stage0_block0"
+ top: "conv_stage0_block1_branch2a"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 32
+ pad: 0
+ kernel_size: 1
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage0_block1_branch2a"
+ type: "BatchNorm"
+ bottom: "conv_stage0_block1_branch2a"
+ top: "conv_stage0_block1_branch2a"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage0_block1_branch2a"
+ type: "Scale"
+ bottom: "conv_stage0_block1_branch2a"
+ top: "conv_stage0_block1_branch2a"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "relu_stage0_block1_branch2a"
+ type: "ReLU"
+ bottom: "conv_stage0_block1_branch2a"
+ top: "conv_stage0_block1_branch2a"
+}
+layer {
+ name: "conv_stage0_block1_branch2b"
+ type: "Convolution"
+ bottom: "conv_stage0_block1_branch2a"
+ top: "conv_stage0_block1_branch2b"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 32
+ pad: 1
+ kernel_size: 3
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage0_block1_branch2b"
+ type: "BatchNorm"
+ bottom: "conv_stage0_block1_branch2b"
+ top: "conv_stage0_block1_branch2b"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage0_block1_branch2b"
+ type: "Scale"
+ bottom: "conv_stage0_block1_branch2b"
+ top: "conv_stage0_block1_branch2b"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "relu_stage0_block1_branch2b"
+ type: "ReLU"
+ bottom: "conv_stage0_block1_branch2b"
+ top: "conv_stage0_block1_branch2b"
+}
+layer {
+ name: "conv_stage0_block1_branch2c"
+ type: "Convolution"
+ bottom: "conv_stage0_block1_branch2b"
+ top: "conv_stage0_block1_branch2c"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 128
+ pad: 0
+ kernel_size: 1
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage0_block1_branch2c"
+ type: "BatchNorm"
+ bottom: "conv_stage0_block1_branch2c"
+ top: "conv_stage0_block1_branch2c"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage0_block1_branch2c"
+ type: "Scale"
+ bottom: "conv_stage0_block1_branch2c"
+ top: "conv_stage0_block1_branch2c"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "eltwise_stage0_block1"
+ type: "Eltwise"
+ bottom: "eltwise_stage0_block0"
+ bottom: "conv_stage0_block1_branch2c"
+ top: "eltwise_stage0_block1"
+}
+layer {
+ name: "relu_stage0_block1"
+ type: "ReLU"
+ bottom: "eltwise_stage0_block1"
+ top: "eltwise_stage0_block1"
+}
+layer {
+ name: "conv_stage0_block2_branch2a"
+ type: "Convolution"
+ bottom: "eltwise_stage0_block1"
+ top: "conv_stage0_block2_branch2a"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 32
+ pad: 0
+ kernel_size: 1
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage0_block2_branch2a"
+ type: "BatchNorm"
+ bottom: "conv_stage0_block2_branch2a"
+ top: "conv_stage0_block2_branch2a"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage0_block2_branch2a"
+ type: "Scale"
+ bottom: "conv_stage0_block2_branch2a"
+ top: "conv_stage0_block2_branch2a"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "relu_stage0_block2_branch2a"
+ type: "ReLU"
+ bottom: "conv_stage0_block2_branch2a"
+ top: "conv_stage0_block2_branch2a"
+}
+layer {
+ name: "conv_stage0_block2_branch2b"
+ type: "Convolution"
+ bottom: "conv_stage0_block2_branch2a"
+ top: "conv_stage0_block2_branch2b"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 32
+ pad: 1
+ kernel_size: 3
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage0_block2_branch2b"
+ type: "BatchNorm"
+ bottom: "conv_stage0_block2_branch2b"
+ top: "conv_stage0_block2_branch2b"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage0_block2_branch2b"
+ type: "Scale"
+ bottom: "conv_stage0_block2_branch2b"
+ top: "conv_stage0_block2_branch2b"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "relu_stage0_block2_branch2b"
+ type: "ReLU"
+ bottom: "conv_stage0_block2_branch2b"
+ top: "conv_stage0_block2_branch2b"
+}
+layer {
+ name: "conv_stage0_block2_branch2c"
+ type: "Convolution"
+ bottom: "conv_stage0_block2_branch2b"
+ top: "conv_stage0_block2_branch2c"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 128
+ pad: 0
+ kernel_size: 1
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage0_block2_branch2c"
+ type: "BatchNorm"
+ bottom: "conv_stage0_block2_branch2c"
+ top: "conv_stage0_block2_branch2c"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage0_block2_branch2c"
+ type: "Scale"
+ bottom: "conv_stage0_block2_branch2c"
+ top: "conv_stage0_block2_branch2c"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "eltwise_stage0_block2"
+ type: "Eltwise"
+ bottom: "eltwise_stage0_block1"
+ bottom: "conv_stage0_block2_branch2c"
+ top: "eltwise_stage0_block2"
+}
+layer {
+ name: "relu_stage0_block2"
+ type: "ReLU"
+ bottom: "eltwise_stage0_block2"
+ top: "eltwise_stage0_block2"
+}
+layer {
+ name: "conv_stage1_block0_proj_shortcut"
+ type: "Convolution"
+ bottom: "eltwise_stage0_block2"
+ top: "conv_stage1_block0_proj_shortcut"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 256
+ pad: 0
+ kernel_size: 1
+ stride: 2
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage1_block0_proj_shortcut"
+ type: "BatchNorm"
+ bottom: "conv_stage1_block0_proj_shortcut"
+ top: "conv_stage1_block0_proj_shortcut"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage1_block0_proj_shortcut"
+ type: "Scale"
+ bottom: "conv_stage1_block0_proj_shortcut"
+ top: "conv_stage1_block0_proj_shortcut"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "conv_stage1_block0_branch2a"
+ type: "Convolution"
+ bottom: "eltwise_stage0_block2"
+ top: "conv_stage1_block0_branch2a"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 64
+ pad: 0
+ kernel_size: 1
+ stride: 2
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage1_block0_branch2a"
+ type: "BatchNorm"
+ bottom: "conv_stage1_block0_branch2a"
+ top: "conv_stage1_block0_branch2a"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage1_block0_branch2a"
+ type: "Scale"
+ bottom: "conv_stage1_block0_branch2a"
+ top: "conv_stage1_block0_branch2a"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "relu_stage1_block0_branch2a"
+ type: "ReLU"
+ bottom: "conv_stage1_block0_branch2a"
+ top: "conv_stage1_block0_branch2a"
+}
+layer {
+ name: "conv_stage1_block0_branch2b"
+ type: "Convolution"
+ bottom: "conv_stage1_block0_branch2a"
+ top: "conv_stage1_block0_branch2b"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 64
+ pad: 1
+ kernel_size: 3
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage1_block0_branch2b"
+ type: "BatchNorm"
+ bottom: "conv_stage1_block0_branch2b"
+ top: "conv_stage1_block0_branch2b"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage1_block0_branch2b"
+ type: "Scale"
+ bottom: "conv_stage1_block0_branch2b"
+ top: "conv_stage1_block0_branch2b"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "relu_stage1_block0_branch2b"
+ type: "ReLU"
+ bottom: "conv_stage1_block0_branch2b"
+ top: "conv_stage1_block0_branch2b"
+}
+layer {
+ name: "conv_stage1_block0_branch2c"
+ type: "Convolution"
+ bottom: "conv_stage1_block0_branch2b"
+ top: "conv_stage1_block0_branch2c"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 256
+ pad: 0
+ kernel_size: 1
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage1_block0_branch2c"
+ type: "BatchNorm"
+ bottom: "conv_stage1_block0_branch2c"
+ top: "conv_stage1_block0_branch2c"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage1_block0_branch2c"
+ type: "Scale"
+ bottom: "conv_stage1_block0_branch2c"
+ top: "conv_stage1_block0_branch2c"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "eltwise_stage1_block0"
+ type: "Eltwise"
+ bottom: "conv_stage1_block0_proj_shortcut"
+ bottom: "conv_stage1_block0_branch2c"
+ top: "eltwise_stage1_block0"
+}
+layer {
+ name: "relu_stage1_block0"
+ type: "ReLU"
+ bottom: "eltwise_stage1_block0"
+ top: "eltwise_stage1_block0"
+}
+layer {
+ name: "conv_stage1_block1_branch2a"
+ type: "Convolution"
+ bottom: "eltwise_stage1_block0"
+ top: "conv_stage1_block1_branch2a"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 64
+ pad: 0
+ kernel_size: 1
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage1_block1_branch2a"
+ type: "BatchNorm"
+ bottom: "conv_stage1_block1_branch2a"
+ top: "conv_stage1_block1_branch2a"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage1_block1_branch2a"
+ type: "Scale"
+ bottom: "conv_stage1_block1_branch2a"
+ top: "conv_stage1_block1_branch2a"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "relu_stage1_block1_branch2a"
+ type: "ReLU"
+ bottom: "conv_stage1_block1_branch2a"
+ top: "conv_stage1_block1_branch2a"
+}
+layer {
+ name: "conv_stage1_block1_branch2b"
+ type: "Convolution"
+ bottom: "conv_stage1_block1_branch2a"
+ top: "conv_stage1_block1_branch2b"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 64
+ pad: 1
+ kernel_size: 3
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage1_block1_branch2b"
+ type: "BatchNorm"
+ bottom: "conv_stage1_block1_branch2b"
+ top: "conv_stage1_block1_branch2b"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage1_block1_branch2b"
+ type: "Scale"
+ bottom: "conv_stage1_block1_branch2b"
+ top: "conv_stage1_block1_branch2b"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "relu_stage1_block1_branch2b"
+ type: "ReLU"
+ bottom: "conv_stage1_block1_branch2b"
+ top: "conv_stage1_block1_branch2b"
+}
+layer {
+ name: "conv_stage1_block1_branch2c"
+ type: "Convolution"
+ bottom: "conv_stage1_block1_branch2b"
+ top: "conv_stage1_block1_branch2c"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 256
+ pad: 0
+ kernel_size: 1
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage1_block1_branch2c"
+ type: "BatchNorm"
+ bottom: "conv_stage1_block1_branch2c"
+ top: "conv_stage1_block1_branch2c"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage1_block1_branch2c"
+ type: "Scale"
+ bottom: "conv_stage1_block1_branch2c"
+ top: "conv_stage1_block1_branch2c"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "eltwise_stage1_block1"
+ type: "Eltwise"
+ bottom: "eltwise_stage1_block0"
+ bottom: "conv_stage1_block1_branch2c"
+ top: "eltwise_stage1_block1"
+}
+layer {
+ name: "relu_stage1_block1"
+ type: "ReLU"
+ bottom: "eltwise_stage1_block1"
+ top: "eltwise_stage1_block1"
+}
+layer {
+ name: "conv_stage1_block2_branch2a"
+ type: "Convolution"
+ bottom: "eltwise_stage1_block1"
+ top: "conv_stage1_block2_branch2a"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 64
+ pad: 0
+ kernel_size: 1
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage1_block2_branch2a"
+ type: "BatchNorm"
+ bottom: "conv_stage1_block2_branch2a"
+ top: "conv_stage1_block2_branch2a"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage1_block2_branch2a"
+ type: "Scale"
+ bottom: "conv_stage1_block2_branch2a"
+ top: "conv_stage1_block2_branch2a"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "relu_stage1_block2_branch2a"
+ type: "ReLU"
+ bottom: "conv_stage1_block2_branch2a"
+ top: "conv_stage1_block2_branch2a"
+}
+layer {
+ name: "conv_stage1_block2_branch2b"
+ type: "Convolution"
+ bottom: "conv_stage1_block2_branch2a"
+ top: "conv_stage1_block2_branch2b"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 64
+ pad: 1
+ kernel_size: 3
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage1_block2_branch2b"
+ type: "BatchNorm"
+ bottom: "conv_stage1_block2_branch2b"
+ top: "conv_stage1_block2_branch2b"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage1_block2_branch2b"
+ type: "Scale"
+ bottom: "conv_stage1_block2_branch2b"
+ top: "conv_stage1_block2_branch2b"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "relu_stage1_block2_branch2b"
+ type: "ReLU"
+ bottom: "conv_stage1_block2_branch2b"
+ top: "conv_stage1_block2_branch2b"
+}
+layer {
+ name: "conv_stage1_block2_branch2c"
+ type: "Convolution"
+ bottom: "conv_stage1_block2_branch2b"
+ top: "conv_stage1_block2_branch2c"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 256
+ pad: 0
+ kernel_size: 1
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage1_block2_branch2c"
+ type: "BatchNorm"
+ bottom: "conv_stage1_block2_branch2c"
+ top: "conv_stage1_block2_branch2c"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage1_block2_branch2c"
+ type: "Scale"
+ bottom: "conv_stage1_block2_branch2c"
+ top: "conv_stage1_block2_branch2c"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "eltwise_stage1_block2"
+ type: "Eltwise"
+ bottom: "eltwise_stage1_block1"
+ bottom: "conv_stage1_block2_branch2c"
+ top: "eltwise_stage1_block2"
+}
+layer {
+ name: "relu_stage1_block2"
+ type: "ReLU"
+ bottom: "eltwise_stage1_block2"
+ top: "eltwise_stage1_block2"
+}
+layer {
+ name: "conv_stage1_block3_branch2a"
+ type: "Convolution"
+ bottom: "eltwise_stage1_block2"
+ top: "conv_stage1_block3_branch2a"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 64
+ pad: 0
+ kernel_size: 1
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage1_block3_branch2a"
+ type: "BatchNorm"
+ bottom: "conv_stage1_block3_branch2a"
+ top: "conv_stage1_block3_branch2a"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage1_block3_branch2a"
+ type: "Scale"
+ bottom: "conv_stage1_block3_branch2a"
+ top: "conv_stage1_block3_branch2a"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "relu_stage1_block3_branch2a"
+ type: "ReLU"
+ bottom: "conv_stage1_block3_branch2a"
+ top: "conv_stage1_block3_branch2a"
+}
+layer {
+ name: "conv_stage1_block3_branch2b"
+ type: "Convolution"
+ bottom: "conv_stage1_block3_branch2a"
+ top: "conv_stage1_block3_branch2b"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 64
+ pad: 1
+ kernel_size: 3
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage1_block3_branch2b"
+ type: "BatchNorm"
+ bottom: "conv_stage1_block3_branch2b"
+ top: "conv_stage1_block3_branch2b"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage1_block3_branch2b"
+ type: "Scale"
+ bottom: "conv_stage1_block3_branch2b"
+ top: "conv_stage1_block3_branch2b"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "relu_stage1_block3_branch2b"
+ type: "ReLU"
+ bottom: "conv_stage1_block3_branch2b"
+ top: "conv_stage1_block3_branch2b"
+}
+layer {
+ name: "conv_stage1_block3_branch2c"
+ type: "Convolution"
+ bottom: "conv_stage1_block3_branch2b"
+ top: "conv_stage1_block3_branch2c"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 256
+ pad: 0
+ kernel_size: 1
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage1_block3_branch2c"
+ type: "BatchNorm"
+ bottom: "conv_stage1_block3_branch2c"
+ top: "conv_stage1_block3_branch2c"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage1_block3_branch2c"
+ type: "Scale"
+ bottom: "conv_stage1_block3_branch2c"
+ top: "conv_stage1_block3_branch2c"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "eltwise_stage1_block3"
+ type: "Eltwise"
+ bottom: "eltwise_stage1_block2"
+ bottom: "conv_stage1_block3_branch2c"
+ top: "eltwise_stage1_block3"
+}
+layer {
+ name: "relu_stage1_block3"
+ type: "ReLU"
+ bottom: "eltwise_stage1_block3"
+ top: "eltwise_stage1_block3"
+}
+layer {
+ name: "conv_stage2_block0_proj_shortcut"
+ type: "Convolution"
+ bottom: "eltwise_stage1_block3"
+ top: "conv_stage2_block0_proj_shortcut"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 512
+ pad: 0
+ kernel_size: 1
+ stride: 2
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage2_block0_proj_shortcut"
+ type: "BatchNorm"
+ bottom: "conv_stage2_block0_proj_shortcut"
+ top: "conv_stage2_block0_proj_shortcut"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage2_block0_proj_shortcut"
+ type: "Scale"
+ bottom: "conv_stage2_block0_proj_shortcut"
+ top: "conv_stage2_block0_proj_shortcut"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "conv_stage2_block0_branch2a"
+ type: "Convolution"
+ bottom: "eltwise_stage1_block3"
+ top: "conv_stage2_block0_branch2a"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 128
+ pad: 0
+ kernel_size: 1
+ stride: 2
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage2_block0_branch2a"
+ type: "BatchNorm"
+ bottom: "conv_stage2_block0_branch2a"
+ top: "conv_stage2_block0_branch2a"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage2_block0_branch2a"
+ type: "Scale"
+ bottom: "conv_stage2_block0_branch2a"
+ top: "conv_stage2_block0_branch2a"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "relu_stage2_block0_branch2a"
+ type: "ReLU"
+ bottom: "conv_stage2_block0_branch2a"
+ top: "conv_stage2_block0_branch2a"
+}
+layer {
+ name: "conv_stage2_block0_branch2b"
+ type: "Convolution"
+ bottom: "conv_stage2_block0_branch2a"
+ top: "conv_stage2_block0_branch2b"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 128
+ pad: 1
+ kernel_size: 3
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage2_block0_branch2b"
+ type: "BatchNorm"
+ bottom: "conv_stage2_block0_branch2b"
+ top: "conv_stage2_block0_branch2b"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage2_block0_branch2b"
+ type: "Scale"
+ bottom: "conv_stage2_block0_branch2b"
+ top: "conv_stage2_block0_branch2b"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "relu_stage2_block0_branch2b"
+ type: "ReLU"
+ bottom: "conv_stage2_block0_branch2b"
+ top: "conv_stage2_block0_branch2b"
+}
+layer {
+ name: "conv_stage2_block0_branch2c"
+ type: "Convolution"
+ bottom: "conv_stage2_block0_branch2b"
+ top: "conv_stage2_block0_branch2c"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 512
+ pad: 0
+ kernel_size: 1
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage2_block0_branch2c"
+ type: "BatchNorm"
+ bottom: "conv_stage2_block0_branch2c"
+ top: "conv_stage2_block0_branch2c"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage2_block0_branch2c"
+ type: "Scale"
+ bottom: "conv_stage2_block0_branch2c"
+ top: "conv_stage2_block0_branch2c"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "eltwise_stage2_block0"
+ type: "Eltwise"
+ bottom: "conv_stage2_block0_proj_shortcut"
+ bottom: "conv_stage2_block0_branch2c"
+ top: "eltwise_stage2_block0"
+}
+layer {
+ name: "relu_stage2_block0"
+ type: "ReLU"
+ bottom: "eltwise_stage2_block0"
+ top: "eltwise_stage2_block0"
+}
+layer {
+ name: "conv_stage2_block1_branch2a"
+ type: "Convolution"
+ bottom: "eltwise_stage2_block0"
+ top: "conv_stage2_block1_branch2a"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 128
+ pad: 0
+ kernel_size: 1
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage2_block1_branch2a"
+ type: "BatchNorm"
+ bottom: "conv_stage2_block1_branch2a"
+ top: "conv_stage2_block1_branch2a"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage2_block1_branch2a"
+ type: "Scale"
+ bottom: "conv_stage2_block1_branch2a"
+ top: "conv_stage2_block1_branch2a"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "relu_stage2_block1_branch2a"
+ type: "ReLU"
+ bottom: "conv_stage2_block1_branch2a"
+ top: "conv_stage2_block1_branch2a"
+}
+layer {
+ name: "conv_stage2_block1_branch2b"
+ type: "Convolution"
+ bottom: "conv_stage2_block1_branch2a"
+ top: "conv_stage2_block1_branch2b"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 128
+ pad: 1
+ kernel_size: 3
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage2_block1_branch2b"
+ type: "BatchNorm"
+ bottom: "conv_stage2_block1_branch2b"
+ top: "conv_stage2_block1_branch2b"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage2_block1_branch2b"
+ type: "Scale"
+ bottom: "conv_stage2_block1_branch2b"
+ top: "conv_stage2_block1_branch2b"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "relu_stage2_block1_branch2b"
+ type: "ReLU"
+ bottom: "conv_stage2_block1_branch2b"
+ top: "conv_stage2_block1_branch2b"
+}
+layer {
+ name: "conv_stage2_block1_branch2c"
+ type: "Convolution"
+ bottom: "conv_stage2_block1_branch2b"
+ top: "conv_stage2_block1_branch2c"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 512
+ pad: 0
+ kernel_size: 1
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage2_block1_branch2c"
+ type: "BatchNorm"
+ bottom: "conv_stage2_block1_branch2c"
+ top: "conv_stage2_block1_branch2c"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage2_block1_branch2c"
+ type: "Scale"
+ bottom: "conv_stage2_block1_branch2c"
+ top: "conv_stage2_block1_branch2c"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "eltwise_stage2_block1"
+ type: "Eltwise"
+ bottom: "eltwise_stage2_block0"
+ bottom: "conv_stage2_block1_branch2c"
+ top: "eltwise_stage2_block1"
+}
+layer {
+ name: "relu_stage2_block1"
+ type: "ReLU"
+ bottom: "eltwise_stage2_block1"
+ top: "eltwise_stage2_block1"
+}
+layer {
+ name: "conv_stage2_block2_branch2a"
+ type: "Convolution"
+ bottom: "eltwise_stage2_block1"
+ top: "conv_stage2_block2_branch2a"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 128
+ pad: 0
+ kernel_size: 1
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage2_block2_branch2a"
+ type: "BatchNorm"
+ bottom: "conv_stage2_block2_branch2a"
+ top: "conv_stage2_block2_branch2a"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage2_block2_branch2a"
+ type: "Scale"
+ bottom: "conv_stage2_block2_branch2a"
+ top: "conv_stage2_block2_branch2a"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "relu_stage2_block2_branch2a"
+ type: "ReLU"
+ bottom: "conv_stage2_block2_branch2a"
+ top: "conv_stage2_block2_branch2a"
+}
+layer {
+ name: "conv_stage2_block2_branch2b"
+ type: "Convolution"
+ bottom: "conv_stage2_block2_branch2a"
+ top: "conv_stage2_block2_branch2b"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 128
+ pad: 1
+ kernel_size: 3
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage2_block2_branch2b"
+ type: "BatchNorm"
+ bottom: "conv_stage2_block2_branch2b"
+ top: "conv_stage2_block2_branch2b"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage2_block2_branch2b"
+ type: "Scale"
+ bottom: "conv_stage2_block2_branch2b"
+ top: "conv_stage2_block2_branch2b"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "relu_stage2_block2_branch2b"
+ type: "ReLU"
+ bottom: "conv_stage2_block2_branch2b"
+ top: "conv_stage2_block2_branch2b"
+}
+layer {
+ name: "conv_stage2_block2_branch2c"
+ type: "Convolution"
+ bottom: "conv_stage2_block2_branch2b"
+ top: "conv_stage2_block2_branch2c"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 512
+ pad: 0
+ kernel_size: 1
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage2_block2_branch2c"
+ type: "BatchNorm"
+ bottom: "conv_stage2_block2_branch2c"
+ top: "conv_stage2_block2_branch2c"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage2_block2_branch2c"
+ type: "Scale"
+ bottom: "conv_stage2_block2_branch2c"
+ top: "conv_stage2_block2_branch2c"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "eltwise_stage2_block2"
+ type: "Eltwise"
+ bottom: "eltwise_stage2_block1"
+ bottom: "conv_stage2_block2_branch2c"
+ top: "eltwise_stage2_block2"
+}
+layer {
+ name: "relu_stage2_block2"
+ type: "ReLU"
+ bottom: "eltwise_stage2_block2"
+ top: "eltwise_stage2_block2"
+}
+layer {
+ name: "conv_stage2_block3_branch2a"
+ type: "Convolution"
+ bottom: "eltwise_stage2_block2"
+ top: "conv_stage2_block3_branch2a"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 128
+ pad: 0
+ kernel_size: 1
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage2_block3_branch2a"
+ type: "BatchNorm"
+ bottom: "conv_stage2_block3_branch2a"
+ top: "conv_stage2_block3_branch2a"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage2_block3_branch2a"
+ type: "Scale"
+ bottom: "conv_stage2_block3_branch2a"
+ top: "conv_stage2_block3_branch2a"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "relu_stage2_block3_branch2a"
+ type: "ReLU"
+ bottom: "conv_stage2_block3_branch2a"
+ top: "conv_stage2_block3_branch2a"
+}
+layer {
+ name: "conv_stage2_block3_branch2b"
+ type: "Convolution"
+ bottom: "conv_stage2_block3_branch2a"
+ top: "conv_stage2_block3_branch2b"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 128
+ pad: 1
+ kernel_size: 3
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage2_block3_branch2b"
+ type: "BatchNorm"
+ bottom: "conv_stage2_block3_branch2b"
+ top: "conv_stage2_block3_branch2b"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage2_block3_branch2b"
+ type: "Scale"
+ bottom: "conv_stage2_block3_branch2b"
+ top: "conv_stage2_block3_branch2b"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "relu_stage2_block3_branch2b"
+ type: "ReLU"
+ bottom: "conv_stage2_block3_branch2b"
+ top: "conv_stage2_block3_branch2b"
+}
+layer {
+ name: "conv_stage2_block3_branch2c"
+ type: "Convolution"
+ bottom: "conv_stage2_block3_branch2b"
+ top: "conv_stage2_block3_branch2c"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 512
+ pad: 0
+ kernel_size: 1
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage2_block3_branch2c"
+ type: "BatchNorm"
+ bottom: "conv_stage2_block3_branch2c"
+ top: "conv_stage2_block3_branch2c"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage2_block3_branch2c"
+ type: "Scale"
+ bottom: "conv_stage2_block3_branch2c"
+ top: "conv_stage2_block3_branch2c"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "eltwise_stage2_block3"
+ type: "Eltwise"
+ bottom: "eltwise_stage2_block2"
+ bottom: "conv_stage2_block3_branch2c"
+ top: "eltwise_stage2_block3"
+}
+layer {
+ name: "relu_stage2_block3"
+ type: "ReLU"
+ bottom: "eltwise_stage2_block3"
+ top: "eltwise_stage2_block3"
+}
+layer {
+ name: "conv_stage2_block4_branch2a"
+ type: "Convolution"
+ bottom: "eltwise_stage2_block3"
+ top: "conv_stage2_block4_branch2a"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 128
+ pad: 0
+ kernel_size: 1
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage2_block4_branch2a"
+ type: "BatchNorm"
+ bottom: "conv_stage2_block4_branch2a"
+ top: "conv_stage2_block4_branch2a"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage2_block4_branch2a"
+ type: "Scale"
+ bottom: "conv_stage2_block4_branch2a"
+ top: "conv_stage2_block4_branch2a"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "relu_stage2_block4_branch2a"
+ type: "ReLU"
+ bottom: "conv_stage2_block4_branch2a"
+ top: "conv_stage2_block4_branch2a"
+}
+layer {
+ name: "conv_stage2_block4_branch2b"
+ type: "Convolution"
+ bottom: "conv_stage2_block4_branch2a"
+ top: "conv_stage2_block4_branch2b"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 128
+ pad: 1
+ kernel_size: 3
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage2_block4_branch2b"
+ type: "BatchNorm"
+ bottom: "conv_stage2_block4_branch2b"
+ top: "conv_stage2_block4_branch2b"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage2_block4_branch2b"
+ type: "Scale"
+ bottom: "conv_stage2_block4_branch2b"
+ top: "conv_stage2_block4_branch2b"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "relu_stage2_block4_branch2b"
+ type: "ReLU"
+ bottom: "conv_stage2_block4_branch2b"
+ top: "conv_stage2_block4_branch2b"
+}
+layer {
+ name: "conv_stage2_block4_branch2c"
+ type: "Convolution"
+ bottom: "conv_stage2_block4_branch2b"
+ top: "conv_stage2_block4_branch2c"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 512
+ pad: 0
+ kernel_size: 1
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage2_block4_branch2c"
+ type: "BatchNorm"
+ bottom: "conv_stage2_block4_branch2c"
+ top: "conv_stage2_block4_branch2c"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage2_block4_branch2c"
+ type: "Scale"
+ bottom: "conv_stage2_block4_branch2c"
+ top: "conv_stage2_block4_branch2c"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "eltwise_stage2_block4"
+ type: "Eltwise"
+ bottom: "eltwise_stage2_block3"
+ bottom: "conv_stage2_block4_branch2c"
+ top: "eltwise_stage2_block4"
+}
+layer {
+ name: "relu_stage2_block4"
+ type: "ReLU"
+ bottom: "eltwise_stage2_block4"
+ top: "eltwise_stage2_block4"
+}
+layer {
+ name: "conv_stage2_block5_branch2a"
+ type: "Convolution"
+ bottom: "eltwise_stage2_block4"
+ top: "conv_stage2_block5_branch2a"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 128
+ pad: 0
+ kernel_size: 1
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage2_block5_branch2a"
+ type: "BatchNorm"
+ bottom: "conv_stage2_block5_branch2a"
+ top: "conv_stage2_block5_branch2a"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage2_block5_branch2a"
+ type: "Scale"
+ bottom: "conv_stage2_block5_branch2a"
+ top: "conv_stage2_block5_branch2a"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "relu_stage2_block5_branch2a"
+ type: "ReLU"
+ bottom: "conv_stage2_block5_branch2a"
+ top: "conv_stage2_block5_branch2a"
+}
+layer {
+ name: "conv_stage2_block5_branch2b"
+ type: "Convolution"
+ bottom: "conv_stage2_block5_branch2a"
+ top: "conv_stage2_block5_branch2b"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 128
+ pad: 1
+ kernel_size: 3
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage2_block5_branch2b"
+ type: "BatchNorm"
+ bottom: "conv_stage2_block5_branch2b"
+ top: "conv_stage2_block5_branch2b"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage2_block5_branch2b"
+ type: "Scale"
+ bottom: "conv_stage2_block5_branch2b"
+ top: "conv_stage2_block5_branch2b"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "relu_stage2_block5_branch2b"
+ type: "ReLU"
+ bottom: "conv_stage2_block5_branch2b"
+ top: "conv_stage2_block5_branch2b"
+}
+layer {
+ name: "conv_stage2_block5_branch2c"
+ type: "Convolution"
+ bottom: "conv_stage2_block5_branch2b"
+ top: "conv_stage2_block5_branch2c"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 512
+ pad: 0
+ kernel_size: 1
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage2_block5_branch2c"
+ type: "BatchNorm"
+ bottom: "conv_stage2_block5_branch2c"
+ top: "conv_stage2_block5_branch2c"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage2_block5_branch2c"
+ type: "Scale"
+ bottom: "conv_stage2_block5_branch2c"
+ top: "conv_stage2_block5_branch2c"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "eltwise_stage2_block5"
+ type: "Eltwise"
+ bottom: "eltwise_stage2_block4"
+ bottom: "conv_stage2_block5_branch2c"
+ top: "eltwise_stage2_block5"
+}
+layer {
+ name: "relu_stage2_block5"
+ type: "ReLU"
+ bottom: "eltwise_stage2_block5"
+ top: "eltwise_stage2_block5"
+}
+layer {
+ name: "conv_stage3_block0_proj_shortcut"
+ type: "Convolution"
+ bottom: "eltwise_stage2_block5"
+ top: "conv_stage3_block0_proj_shortcut"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 1024
+ pad: 0
+ kernel_size: 1
+ stride: 2
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage3_block0_proj_shortcut"
+ type: "BatchNorm"
+ bottom: "conv_stage3_block0_proj_shortcut"
+ top: "conv_stage3_block0_proj_shortcut"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage3_block0_proj_shortcut"
+ type: "Scale"
+ bottom: "conv_stage3_block0_proj_shortcut"
+ top: "conv_stage3_block0_proj_shortcut"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "conv_stage3_block0_branch2a"
+ type: "Convolution"
+ bottom: "eltwise_stage2_block5"
+ top: "conv_stage3_block0_branch2a"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 256
+ pad: 0
+ kernel_size: 1
+ stride: 2
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage3_block0_branch2a"
+ type: "BatchNorm"
+ bottom: "conv_stage3_block0_branch2a"
+ top: "conv_stage3_block0_branch2a"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage3_block0_branch2a"
+ type: "Scale"
+ bottom: "conv_stage3_block0_branch2a"
+ top: "conv_stage3_block0_branch2a"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "relu_stage3_block0_branch2a"
+ type: "ReLU"
+ bottom: "conv_stage3_block0_branch2a"
+ top: "conv_stage3_block0_branch2a"
+}
+layer {
+ name: "conv_stage3_block0_branch2b"
+ type: "Convolution"
+ bottom: "conv_stage3_block0_branch2a"
+ top: "conv_stage3_block0_branch2b"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 256
+ pad: 1
+ kernel_size: 3
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage3_block0_branch2b"
+ type: "BatchNorm"
+ bottom: "conv_stage3_block0_branch2b"
+ top: "conv_stage3_block0_branch2b"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage3_block0_branch2b"
+ type: "Scale"
+ bottom: "conv_stage3_block0_branch2b"
+ top: "conv_stage3_block0_branch2b"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "relu_stage3_block0_branch2b"
+ type: "ReLU"
+ bottom: "conv_stage3_block0_branch2b"
+ top: "conv_stage3_block0_branch2b"
+}
+layer {
+ name: "conv_stage3_block0_branch2c"
+ type: "Convolution"
+ bottom: "conv_stage3_block0_branch2b"
+ top: "conv_stage3_block0_branch2c"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 1024
+ pad: 0
+ kernel_size: 1
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage3_block0_branch2c"
+ type: "BatchNorm"
+ bottom: "conv_stage3_block0_branch2c"
+ top: "conv_stage3_block0_branch2c"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage3_block0_branch2c"
+ type: "Scale"
+ bottom: "conv_stage3_block0_branch2c"
+ top: "conv_stage3_block0_branch2c"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "eltwise_stage3_block0"
+ type: "Eltwise"
+ bottom: "conv_stage3_block0_proj_shortcut"
+ bottom: "conv_stage3_block0_branch2c"
+ top: "eltwise_stage3_block0"
+}
+layer {
+ name: "relu_stage3_block0"
+ type: "ReLU"
+ bottom: "eltwise_stage3_block0"
+ top: "eltwise_stage3_block0"
+}
+layer {
+ name: "conv_stage3_block1_branch2a"
+ type: "Convolution"
+ bottom: "eltwise_stage3_block0"
+ top: "conv_stage3_block1_branch2a"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 256
+ pad: 0
+ kernel_size: 1
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage3_block1_branch2a"
+ type: "BatchNorm"
+ bottom: "conv_stage3_block1_branch2a"
+ top: "conv_stage3_block1_branch2a"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage3_block1_branch2a"
+ type: "Scale"
+ bottom: "conv_stage3_block1_branch2a"
+ top: "conv_stage3_block1_branch2a"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "relu_stage3_block1_branch2a"
+ type: "ReLU"
+ bottom: "conv_stage3_block1_branch2a"
+ top: "conv_stage3_block1_branch2a"
+}
+layer {
+ name: "conv_stage3_block1_branch2b"
+ type: "Convolution"
+ bottom: "conv_stage3_block1_branch2a"
+ top: "conv_stage3_block1_branch2b"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 256
+ pad: 1
+ kernel_size: 3
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage3_block1_branch2b"
+ type: "BatchNorm"
+ bottom: "conv_stage3_block1_branch2b"
+ top: "conv_stage3_block1_branch2b"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage3_block1_branch2b"
+ type: "Scale"
+ bottom: "conv_stage3_block1_branch2b"
+ top: "conv_stage3_block1_branch2b"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "relu_stage3_block1_branch2b"
+ type: "ReLU"
+ bottom: "conv_stage3_block1_branch2b"
+ top: "conv_stage3_block1_branch2b"
+}
+layer {
+ name: "conv_stage3_block1_branch2c"
+ type: "Convolution"
+ bottom: "conv_stage3_block1_branch2b"
+ top: "conv_stage3_block1_branch2c"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 1024
+ pad: 0
+ kernel_size: 1
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage3_block1_branch2c"
+ type: "BatchNorm"
+ bottom: "conv_stage3_block1_branch2c"
+ top: "conv_stage3_block1_branch2c"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage3_block1_branch2c"
+ type: "Scale"
+ bottom: "conv_stage3_block1_branch2c"
+ top: "conv_stage3_block1_branch2c"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "eltwise_stage3_block1"
+ type: "Eltwise"
+ bottom: "eltwise_stage3_block0"
+ bottom: "conv_stage3_block1_branch2c"
+ top: "eltwise_stage3_block1"
+}
+layer {
+ name: "relu_stage3_block1"
+ type: "ReLU"
+ bottom: "eltwise_stage3_block1"
+ top: "eltwise_stage3_block1"
+}
+layer {
+ name: "conv_stage3_block2_branch2a"
+ type: "Convolution"
+ bottom: "eltwise_stage3_block1"
+ top: "conv_stage3_block2_branch2a"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 256
+ pad: 0
+ kernel_size: 1
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage3_block2_branch2a"
+ type: "BatchNorm"
+ bottom: "conv_stage3_block2_branch2a"
+ top: "conv_stage3_block2_branch2a"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage3_block2_branch2a"
+ type: "Scale"
+ bottom: "conv_stage3_block2_branch2a"
+ top: "conv_stage3_block2_branch2a"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "relu_stage3_block2_branch2a"
+ type: "ReLU"
+ bottom: "conv_stage3_block2_branch2a"
+ top: "conv_stage3_block2_branch2a"
+}
+layer {
+ name: "conv_stage3_block2_branch2b"
+ type: "Convolution"
+ bottom: "conv_stage3_block2_branch2a"
+ top: "conv_stage3_block2_branch2b"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 256
+ pad: 1
+ kernel_size: 3
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage3_block2_branch2b"
+ type: "BatchNorm"
+ bottom: "conv_stage3_block2_branch2b"
+ top: "conv_stage3_block2_branch2b"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage3_block2_branch2b"
+ type: "Scale"
+ bottom: "conv_stage3_block2_branch2b"
+ top: "conv_stage3_block2_branch2b"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "relu_stage3_block2_branch2b"
+ type: "ReLU"
+ bottom: "conv_stage3_block2_branch2b"
+ top: "conv_stage3_block2_branch2b"
+}
+layer {
+ name: "conv_stage3_block2_branch2c"
+ type: "Convolution"
+ bottom: "conv_stage3_block2_branch2b"
+ top: "conv_stage3_block2_branch2c"
+ param {
+ lr_mult: 1
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 2
+ decay_mult: 0
+ }
+ convolution_param {
+ num_output: 1024
+ pad: 0
+ kernel_size: 1
+ stride: 1
+ weight_filler {
+ type: "xavier"
+ }
+ bias_filler {
+ type: "constant"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "bn_stage3_block2_branch2c"
+ type: "BatchNorm"
+ bottom: "conv_stage3_block2_branch2c"
+ top: "conv_stage3_block2_branch2c"
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ param {
+ lr_mult: 0
+ decay_mult: 0
+ }
+ batch_norm_param {
+ use_global_stats: true
+ }
+}
+layer {
+ name: "scale_stage3_block2_branch2c"
+ type: "Scale"
+ bottom: "conv_stage3_block2_branch2c"
+ top: "conv_stage3_block2_branch2c"
+ scale_param {
+ bias_term: true
+ }
+}
+layer {
+ name: "eltwise_stage3_block2"
+ type: "Eltwise"
+ bottom: "eltwise_stage3_block1"
+ bottom: "conv_stage3_block2_branch2c"
+ top: "eltwise_stage3_block2"
+}
+layer {
+ name: "relu_stage3_block2"
+ type: "ReLU"
+ bottom: "eltwise_stage3_block2"
+ top: "eltwise_stage3_block2"
+}
+layer {
+ name: "pool"
+ type: "Pooling"
+ bottom: "eltwise_stage3_block2"
+ top: "pool"
+ pooling_param {
+ pool: AVE
+ kernel_size: 7
+ stride: 1
+ }
+}
+layer {
+ name: "fc_nsfw"
+ type: "InnerProduct"
+ bottom: "pool"
+ top: "fc_nsfw"
+ param {
+ lr_mult: 5
+ decay_mult: 1
+ }
+ param {
+ lr_mult: 10
+ decay_mult: 0
+ }
+ inner_product_param{
+ num_output: 2
+ weight_filler {
+ type: "xavier"
+ std: 0.01
+ }
+ bias_filler {
+ type: "xavier"
+ value: 0
+ }
+ }
+}
+layer {
+ name: "prob"
+ type: "Softmax"
+ bottom: "fc_nsfw"
+ top: "prob"
+}
+
diff --git a/nsfw_model/resnet_50_1by2_nsfw.caffemodel b/nsfw_model/resnet_50_1by2_nsfw.caffemodel
Binary files differ.
