Skip to content

Instantly share code, notes, and snippets.

@helxsz

helxsz/readme.md

Forked from bogger/deploy.prototxt
Last active August 29, 2015 14:27
Show Gist options
  • Save helxsz/ea209a2a82644622bae2 to your computer and use it in GitHub Desktop.
Save helxsz/ea209a2a82644622bae2 to your computer and use it in GitHub Desktop.
Download ZIP
GoogLeNet_cars
Raw
readme.md
name caffemodel caffemodel_url license caffe_commit gist_id
GoogLeNet_cars
googlenet_finetune_web_car_iter_10000.caffemodel
non-commercial
737ea5e936821b5c69f9c3952d72693ae5843370
b90eb88e31cd745525ae

##Description

This is the GoogLeNet model pre-trained on ImageNet classification task and fine-tuned on 431 car models in CompCars dataset. It is described in the technical report. Please cite the following work if the model is useful for you.

A Large-Scale Car Dataset for Fine-Grained Categorization and Verification
L. Yang, P. Luo, C. C. Loy, X. Tang, arXiv:1506.08959, 2015

The bundled model is the iteration 10,000 snapshot. This model obtains a top-1 accuracy 91.2% and a top-5 accuracy 98.1% on the testing set, using only the center crop.

License

The data used to train this model comes from the ImageNet project and the CompCars dataset, which distribute their databases to researchers who agree to a following term of access: "Researcher shall use the Database only for non-commercial research and educational purposes." Accordingly, this model is distributed under a non-commercial license.

Raw
solver_googlenet.prototxt
net: "./train_val_googlenet.prototxt"
test_iter: 500
test_interval: 500
test_initialization: false
max_iter: 10000
base_lr: 0.001
lr_policy: "step"
gamma: 0.1
stepsize: 5000
momentum: 0.9
weight_decay: 0.0002
display: 20
snapshot: 5000
snapshot_prefix: "googlenet_finetune_web_car"
solver_mode: GPU
device_id: 0
random_seed: 34234562302122
Raw
train_val_googlenet.prototxt
name: "GoogleNet"
# train data
layers {
top: "data"
top: "label"
name: "data"
type: IMAGE_DATA
image_data_param {
source: "classification_train"
batch_size: 256
shuffle: true
new_width: 256
new_height: 256
}
transform_param {
mirror: true
crop_size: 224
mean_file: "imagenet_mean.binaryproto"
}
include {
phase: TRAIN
}
}
# test data
layers {
top: "data"
top: "label"
name: "data"
type: IMAGE_DATA
image_data_param {
source: "classification_test"
batch_size: 10
new_width: 256
new_height: 256
}
transform_param {
mirror: false
crop_size: 224
mean_file: "imagenet_mean.binaryproto"
}
include {
phase: TEST
}
}
# hierarchy 1
# conv -> relu -> pool -> lrn
# ===========================
layers {
bottom: "data"
top: "conv1"
name: "conv1"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 64
pad: 3
kernel_size: 7
stride: 2
weight_filler {
type: "xavier"
std: 0.1
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "conv1"
top: "conv1"
name: "relu1"
type: RELU
}
layers {
bottom: "conv1"
top: "pool1"
name: "pool1"
type: POOLING
pooling_param {
pool: MAX
kernel_size: 3
stride: 2
}
}
layers {
bottom: "pool1"
top: "norm1"
name: "norm1"
type: LRN
lrn_param {
local_size: 5
alpha: 0.0001
beta: 0.75
}
}
# hierarchy 2
# conv -> relu -> conv -> relu -> lrn -> pool
# ===========================================
layers {
bottom: "norm1"
top: "conv2_1x1"
name: "conv2_1x1"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 64
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.1
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "conv2_1x1"
top: "conv2_1x1"
name: "relu_conv2_1x1"
type: RELU
}
layers {
bottom: "conv2_1x1"
top: "conv2_3x3"
name: "conv2_3x3"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 192
pad: 1
kernel_size: 3
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "conv2_3x3"
top: "conv2_3x3"
name: "relu2_3x3"
type: RELU
}
layers {
bottom: "conv2_3x3"
top: "norm2"
name: "norm2"
type: LRN
lrn_param {
local_size: 5
alpha: 0.0001
beta: 0.75
}
}
layers {
bottom: "norm2"
top: "pool2"
name: "pool2"
type: POOLING
pooling_param {
pool: MAX
kernel_size: 3
stride: 2
}
}
# ================== hierarchy 3(a) ==================
# ====================================================
# conv 1 x 1
# ----------
layers {
bottom: "pool2"
top: "inception_3a_1x1"
name: "inception_3a_1x1"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 64
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_3a_1x1"
top: "inception_3a_1x1"
name: "relu_inception_3a_1x1"
type: RELU
}
# conv 3 x 3
# ----------
layers {
bottom: "pool2"
top: "inception_3a_3x3_reduce"
name: "inception_3a_3x3_reduce"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 96
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.09
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_3a_3x3_reduce"
top: "inception_3a_3x3_reduce"
name: "reulu_inception_3a_3x3_reduce"
type: RELU
}
layers {
bottom: "inception_3a_3x3_reduce"
top: "inception_3a_3x3"
name: "inception_3a_3x3"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 128
pad: 1
kernel_size: 3
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_3a_3x3"
top: "inception_3a_3x3"
name: "relu_inception_3a_3x3"
type: RELU
}
# conv 5 x 5
# ----------
layers {
bottom: "pool2"
top: "inception_3a_5x5_reduce"
name: "inception_3a_5x5_reduce"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 16
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.2
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_3a_5x5_reduce"
top: "inception_3a_5x5_reduce"
name: "relu_inception_3a_5x5_reduce"
type: RELU
}
layers {
bottom: "inception_3a_5x5_reduce"
top: "inception_3a_5x5"
name: "inception_3a_5x5"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 32
pad: 2
kernel_size: 5
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_3a_5x5"
top: "inception_3a_5x5"
name: "relu_inception_3a_5x5"
type: RELU
}
# pool 3 x 3
# ----------
layers {
bottom: "pool2"
top: "inception_3a_pool"
name: "inception_3a_pool"
type: POOLING
pooling_param {
pool: MAX
kernel_size: 3
stride: 1
pad: 1
}
}
layers {
bottom: "inception_3a_pool"
top: "inception_3a_pool_proj"
name: "inception_3a_pool_proj"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 32
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.1
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_3a_pool_proj"
top: "inception_3a_pool_proj"
name: "relu_inception_3a_pool_proj"
type: RELU
}
# concatenate all
# ---------------
layers {
bottom: "inception_3a_1x1"
bottom: "inception_3a_3x3"
bottom: "inception_3a_5x5"
bottom: "inception_3a_pool_proj"
top: "inception_3a_output"
name: "inception_3a_output"
type: CONCAT
}
# ================== hierarchy 3(b) ==================
# ====================================================
# conv 1 x 1
# ----------
layers {
bottom: "inception_3a_output"
top: "inception_3b_1x1"
name: "inception_3b_1x1"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 128
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_3b_1x1"
top: "inception_3b_1x1"
name: "relu_inception_3b_1x1"
type: RELU
}
# conv 3 x 3
# ----------
layers {
bottom: "inception_3a_output"
top: "inception_3b_3x3_reduce"
name: "inception_3b_3x3_reduce"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 128
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.09
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_3b_3x3_reduce"
top: "inception_3b_3x3_reduce"
name: "relu_inception_3b_3x3_reduce"
type: RELU
}
layers {
bottom: "inception_3b_3x3_reduce"
top: "inception_3b_3x3"
name: "inception_3b_3x3"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 192
pad: 1
kernel_size: 3
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_3b_3x3"
top: "inception_3b_3x3"
name: "relu_inception_3b_3x3"
type: RELU
}
# conv 5 x 5
# ----------
layers {
bottom: "inception_3a_output"
top: "inception_3b_5x5_reduce"
name: "inception_3b_5x5_reduce"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 32
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.2
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_3b_5x5_reduce"
top: "inception_3b_5x5_reduce"
name: "relu_inception_3b_5x5_reduce"
type: RELU
}
layers {
bottom: "inception_3b_5x5_reduce"
top: "inception_3b_5x5"
name: "inception_3b_5x5"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 96
pad: 2
kernel_size: 5
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_3b_5x5"
top: "inception_3b_5x5"
name: "relu_inception_3b_5x5"
type: RELU
}
# pool 3 x 3
# ----------
layers {
bottom: "inception_3a_output"
top: "inception_3b_pool"
name: "inception_3b_pool"
type: POOLING
pooling_param {
pool: MAX
kernel_size: 3
stride: 1
pad: 1
}
}
layers {
bottom: "inception_3b_pool"
top: "inception_3b_pool_proj"
name: "inception_3b_pool_proj"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 64
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.1
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_3b_pool_proj"
top: "inception_3b_pool_proj"
name: "relu_inception_3b_pool_proj"
type: RELU
}
# concatenate all
# ---------------
layers {
bottom: "inception_3b_1x1"
bottom: "inception_3b_3x3"
bottom: "inception_3b_5x5"
bottom: "inception_3b_pool_proj"
top: "inception_3b_output"
name: "inception_3b_output"
type: CONCAT
}
layers {
bottom: "inception_3b_output"
top: "pool3"
name: "pool3"
type: POOLING
pooling_param {
pool: MAX
kernel_size: 3
stride: 2
}
}
# ================== hierarchy 4(a) ==================
# ====================================================
# conv 1 x 1
# ----------
layers {
bottom: "pool3"
top: "inception_4a_1x1"
name: "inception_4a_1x1"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 192
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_4a_1x1"
top: "inception_4a_1x1"
name: "relu_inception_4a_1x1"
type: RELU
}
# conv 3 x 3
# ----------
layers {
bottom: "pool3"
top: "inception_4a_3x3_reduce"
name: "inception_4a_3x3_reduce"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 96
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.09
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_4a_3x3_reduce"
top: "inception_4a_3x3_reduce"
name: "relu_inception_4a_3x3_reduce"
type: RELU
}
layers {
bottom: "inception_4a_3x3_reduce"
top: "inception_4a_3x3"
name: "inception_4a_3x3"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 208
pad: 1
kernel_size: 3
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_4a_3x3"
top: "inception_4a_3x3"
name: "relu_inception_4a_3x3"
type: RELU
}
# conv 5 x 5
# ----------
layers {
bottom: "pool3"
top: "inception_4a_5x5_reduce"
name: "inception_4a_5x5_reduce"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 16
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.2
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_4a_5x5_reduce"
top: "inception_4a_5x5_reduce"
name: "relu_inception_4a_5x5_reduce"
type: RELU
}
layers {
bottom: "inception_4a_5x5_reduce"
top: "inception_4a_5x5"
name: "inception_4a_5x5"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 48
pad: 2
kernel_size: 5
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_4a_5x5"
top: "inception_4a_5x5"
name: "relu_inception_4a_5x5"
type: RELU
}
# pool 3 x 3
# ----------
layers {
bottom: "pool3"
top: "inception_4a_pool"
name: "inception_4a_pool"
type: POOLING
pooling_param {
pool: MAX
kernel_size: 3
stride: 1
pad: 1
}
}
layers {
bottom: "inception_4a_pool"
top: "inception_4a_pool_proj"
name: "inception_4a_pool_proj"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 64
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.1
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_4a_pool_proj"
top: "inception_4a_pool_proj"
name: "relu_inception_4a_pool_proj"
type: RELU
}
# concatenate all
# ---------------
layers {
bottom: "inception_4a_1x1"
bottom: "inception_4a_3x3"
bottom: "inception_4a_5x5"
bottom: "inception_4a_pool_proj"
top: "inception_4a_output"
name: "inception_4a_output"
type: CONCAT
}
# ===================== softmax 0 ====================
# ====================================================
# from this point, name of "RELU" in each layer name
# follows default setting.
# ave pool
layers {
bottom: "inception_4a_output"
top: "loss1_ave_pool"
name: "loss1_ave_pool"
type: POOLING
pooling_param {
pool: AVE
kernel_size: 5
stride: 3
}
}
# conv 1 x 1
layers {
bottom: "loss1_ave_pool"
top: "loss1_conv"
name: "loss1_conv"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 128
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.08
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "loss1_conv"
top: "loss1_conv"
name: "loss1_relu_conv"
type: RELU
}
# fc7-like
layers {
bottom: "loss1_conv"
top: "loss1_fc"
name: "loss1_fc"
type: INNER_PRODUCT
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
inner_product_param {
num_output: 1024
weight_filler {
type: "xavier"
std: 0.02
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "loss1_fc"
top: "loss1_fc"
name: "loss1_relu_fc"
type: RELU
}
layers {
bottom: "loss1_fc"
top: "loss1_fc"
name: "loss1_drop_fc"
type: DROPOUT
dropout_param {
dropout_ratio: 0.7
}
}
# fc8-like
layers {
bottom: "loss1_fc"
top: "loss1_classifier_model"
name: "loss1_classifier_model"
type: INNER_PRODUCT
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
inner_product_param {
num_output: 431
weight_filler {
type: "xavier"
std: 0.0009765625
}
bias_filler {
type: "constant"
value: 0
}
}
}
# loss
layers {
bottom: "loss1_classifier_model"
bottom: "label"
top: "loss1"
name: "loss1"
type: SOFTMAX_LOSS
loss_weight: 0.3
}
# accuracy, top1 and top5
layers {
bottom: "loss1_classifier_model"
bottom: "label"
top: "loss1_top-1"
name: "loss1_top-1"
type: ACCURACY
include {
phase: TEST
}
}
layers {
bottom: "loss1_classifier_model"
bottom: "label"
top: "loss1_top-5"
name: "loss1_top-5"
type: ACCURACY
accuracy_param {
top_k: 5
}
include {
phase: TEST
}
}
# ================== hierarchy 4(b) ==================
# ====================================================
# conv 1 x 1
# ----------
layers {
bottom: "inception_4a_output"
top: "inception_4b_1x1"
name: "inception_4b_1x1"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 160
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_4b_1x1"
top: "inception_4b_1x1"
name: "inception_4b_relu_1x1"
type: RELU
}
# conv 3 x 3
# ----------
layers {
bottom: "inception_4a_output"
top: "inception_4b_3x3_reduce"
name: "inception_4b_3x3_reduce"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 112
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.09
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_4b_3x3_reduce"
top: "inception_4b_3x3_reduce"
name: "inception_4b_relu_3x3_reduce"
type: RELU
}
layers {
bottom: "inception_4b_3x3_reduce"
top: "inception_4b_3x3"
name: "inception_4b_3x3"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 224
pad: 1
kernel_size: 3
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_4b_3x3"
top: "inception_4b_3x3"
name: "inception_4b_relu_3x3"
type: RELU
}
# conv 5 x 5
# ----------
layers {
bottom: "inception_4a_output"
top: "inception_4b_5x5_reduce"
name: "inception_4b_5x5_reduce"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 24
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.2
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_4b_5x5_reduce"
top: "inception_4b_5x5_reduce"
name: "inception_4b_relu_5x5_reduce"
type: RELU
}
layers {
bottom: "inception_4b_5x5_reduce"
top: "inception_4b_5x5"
name: "inception_4b_5x5"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 64
pad: 2
kernel_size: 5
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_4b_5x5"
top: "inception_4b_5x5"
name: "inception_4b_relu_5x5"
type: RELU
}
# pool 3 x 3
# ----------
layers {
bottom: "inception_4a_output"
top: "inception_4b_pool"
name: "inception_4b_pool"
type: POOLING
pooling_param {
pool: MAX
kernel_size: 3
stride: 1
pad: 1
}
}
layers {
bottom: "inception_4b_pool"
top: "inception_4b_pool_proj"
name: "inception_4b_pool_proj"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 64
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.1
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_4b_pool_proj"
top: "inception_4b_pool_proj"
name: "inception_4b_relu_pool_proj"
type: RELU
}
# concatenate all
# ---------------
layers {
bottom: "inception_4b_1x1"
bottom: "inception_4b_3x3"
bottom: "inception_4b_5x5"
bottom: "inception_4b_pool_proj"
top: "inception_4b_output"
name: "inception_4b_output"
type: CONCAT
}
# ================== hierarchy 4(c) ==================
# ====================================================
# conv 1 x 1
# ----------
layers {
bottom: "inception_4b_output"
top: "inception_4c_1x1"
name: "inception_4c_1x1"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 128
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_4c_1x1"
top: "inception_4c_1x1"
name: "inception_4c_relu_1x1"
type: RELU
}
# conv 3 x 3
# ----------
layers {
bottom: "inception_4b_output"
top: "inception_4c_3x3_reduce"
name: "inception_4c_3x3_reduce"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 128
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.09
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_4c_3x3_reduce"
top: "inception_4c_3x3_reduce"
name: "inception_4c_relu_3x3_reduce"
type: RELU
}
layers {
bottom: "inception_4c_3x3_reduce"
top: "inception_4c_3x3"
name: "inception_4c_3x3"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 256
pad: 1
kernel_size: 3
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_4c_3x3"
top: "inception_4c_3x3"
name: "inception_4c_relu_3x3"
type: RELU
}
# conv 5 x 5
# ----------
layers {
bottom: "inception_4b_output"
top: "inception_4c_5x5_reduce"
name: "inception_4c_5x5_reduce"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 24
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.2
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_4c_5x5_reduce"
top: "inception_4c_5x5_reduce"
name: "inception_4c_relu_5x5_reduce"
type: RELU
}
layers {
bottom: "inception_4c_5x5_reduce"
top: "inception_4c_5x5"
name: "inception_4c_5x5"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 64
pad: 2
kernel_size: 5
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_4c_5x5"
top: "inception_4c_5x5"
name: "inception_4c_relu_5x5"
type: RELU
}
# pool 3 x 3
# ----------
layers {
bottom: "inception_4b_output"
top: "inception_4c_pool"
name: "inception_4c_pool"
type: POOLING
pooling_param {
pool: MAX
kernel_size: 3
stride: 1
pad: 1
}
}
layers {
bottom: "inception_4c_pool"
top: "inception_4c_pool_proj"
name: "inception_4c_pool_proj"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 64
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.1
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_4c_pool_proj"
top: "inception_4c_pool_proj"
name: "inception_4c_relu_pool_proj"
type: RELU
}
# concatenate them all
# --------------------
layers {
bottom: "inception_4c_1x1"
bottom: "inception_4c_3x3"
bottom: "inception_4c_5x5"
bottom: "inception_4c_pool_proj"
top: "inception_4c_output"
name: "inception_4c_output"
type: CONCAT
}
# ================== hierarchy 4(d) ==================
# ====================================================
# conv 1 x 1
# ----------
layers {
bottom: "inception_4c_output"
top: "inception_4d_1x1"
name: "inception_4d_1x1"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 112
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_4d_1x1"
top: "inception_4d_1x1"
name: "inception_4d_relu_1x1"
type: RELU
}
# conv 3 x 3
# ----------
layers {
bottom: "inception_4c_output"
top: "inception_4d_3x3_reduce"
name: "inception_4d_3x3_reduce"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 144
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.09
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_4d_3x3_reduce"
top: "inception_4d_3x3_reduce"
name: "inception_4d_relu_3x3_reduce"
type: RELU
}
layers {
bottom: "inception_4d_3x3_reduce"
top: "inception_4d_3x3"
name: "inception_4d_3x3"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 288
pad: 1
kernel_size: 3
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_4d_3x3"
top: "inception_4d_3x3"
name: "inception_4d_relu_3x3"
type: RELU
}
# conv 5 x 5
# ----------
layers {
bottom: "inception_4c_output"
top: "inception_4d_5x5_reduce"
name: "inception_4d_5x5_reduce"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 32
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.2
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_4d_5x5_reduce"
top: "inception_4d_5x5_reduce"
name: "inception_4d_relu_5x5_reduce"
type: RELU
}
layers {
bottom: "inception_4d_5x5_reduce"
top: "inception_4d_5x5"
name: "inception_4d_5x5"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 64
pad: 2
kernel_size: 5
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_4d_5x5"
top: "inception_4d_5x5"
name: "inception_4d_relu_5x5"
type: RELU
}
# pool 3 x 3
# ----------
layers {
bottom: "inception_4c_output"
top: "inception_4d_pool"
name: "inception_4d_pool"
type: POOLING
pooling_param {
pool: MAX
kernel_size: 3
stride: 1
pad: 1
}
}
layers {
bottom: "inception_4d_pool"
top: "inception_4d_pool_proj"
name: "inception_4d_pool_proj"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 64
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.1
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_4d_pool_proj"
top: "inception_4d_pool_proj"
name: "inception_4d_relu_pool_proj"
type: RELU
}
# concatenate them all
# --------------------
layers {
bottom: "inception_4d_1x1"
bottom: "inception_4d_3x3"
bottom: "inception_4d_5x5"
bottom: "inception_4d_pool_proj"
top: "inception_4d_output"
name: "inception_4d_output"
type: CONCAT
}
# ===================== softmax 1 ====================
# ====================================================
# ave_pool -> conv(relu) -> fc -> fc -> loss, accuracy
layers {
bottom: "inception_4d_output"
top: "loss2_ave_pool"
name: "loss2_ave_pool"
type: POOLING
pooling_param {
pool: AVE
kernel_size: 5
stride: 3
}
}
layers {
bottom: "loss2_ave_pool"
top: "loss2_conv"
name: "loss2_conv"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 128
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.08
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "loss2_conv"
top: "loss2_conv"
name: "loss2_relu_conv"
type: RELU
}
layers {
bottom: "loss2_conv"
top: "loss2_fc"
name: "loss2_fc"
type: INNER_PRODUCT
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
inner_product_param {
num_output: 1024
weight_filler {
type: "xavier"
std: 0.02
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "loss2_fc"
top: "loss2_fc"
name: "loss2_relu_fc"
type: RELU
}
layers {
bottom: "loss2_fc"
top: "loss2_fc"
name: "loss2_drop_fc"
type: DROPOUT
dropout_param {
dropout_ratio: 0.7
}
}
layers {
bottom: "loss2_fc"
top: "loss2_classifier_model"
name: "loss2_classifier_model"
type: INNER_PRODUCT
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
inner_product_param {
num_output: 431
weight_filler {
type: "xavier"
std: 0.0009765625
}
bias_filler {
type: "constant"
value: 0
}
}
}
# loss
layers {
bottom: "loss2_classifier_model"
bottom: "label"
top: "loss2"
name: "loss2"
type: SOFTMAX_LOSS
loss_weight: 0.3
}
# accuracy
layers {
bottom: "loss2_classifier_model"
bottom: "label"
top: "loss2_top-1"
name: "loss2_top-1"
type: ACCURACY
include {
phase: TEST
}
}
layers {
bottom: "loss2_classifier_model"
bottom: "label"
top: "loss2_top-5"
name: "loss2_top-5"
type: ACCURACY
accuracy_param {
top_k: 5
}
include {
phase: TEST
}
}
# ================== hierarchy 4(e) ==================
# ====================================================
# conv 1 x 1
# ----------
layers {
bottom: "inception_4d_output"
top: "inception_4e_1x1"
name: "inception_4e_1x1"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 256
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_4e_1x1"
top: "inception_4e_1x1"
name: "inception_4e_relu_1x1"
type: RELU
}
# conv 3 x 3
# ----------
layers {
bottom: "inception_4d_output"
top: "inception_4e_3x3_reduce"
name: "inception_4e_3x3_reduce"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 160
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.09
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_4e_3x3_reduce"
top: "inception_4e_3x3_reduce"
name: "inception_4e_relu_3x3_reduce"
type: RELU
}
layers {
bottom: "inception_4e_3x3_reduce"
top: "inception_4e_3x3"
name: "inception_4e_3x3"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 320
pad: 1
kernel_size: 3
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_4e_3x3"
top: "inception_4e_3x3"
name: "inception_4e_relu_3x3"
type: RELU
}
# conv 5 x 5
# ----------
layers {
bottom: "inception_4d_output"
top: "inception_4e_5x5_reduce"
name: "inception_4e_5x5_reduce"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 32
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.2
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_4e_5x5_reduce"
top: "inception_4e_5x5_reduce"
name: "inception_4e_relu_5x5_reduce"
type: RELU
}
layers {
bottom: "inception_4e_5x5_reduce"
top: "inception_4e_5x5"
name: "inception_4e_5x5"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 128
pad: 2
kernel_size: 5
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_4e_5x5"
top: "inception_4e_5x5"
name: "inception_4e_relu_5x5"
type: RELU
}
# pool 3 x 3
# ----------
layers {
bottom: "inception_4d_output"
top: "inception_4e_pool"
name: "inception_4e_pool"
type: POOLING
pooling_param {
pool: MAX
kernel_size: 3
stride: 1
pad: 1
}
}
layers {
bottom: "inception_4e_pool"
top: "inception_4e_pool_proj"
name: "inception_4e_pool_proj"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 128
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.1
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_4e_pool_proj"
top: "inception_4e_pool_proj"
name: "inception_4e_relu_pool_proj"
type: RELU
}
# concatenate them all
# --------------------
layers {
bottom: "inception_4e_1x1"
bottom: "inception_4e_3x3"
bottom: "inception_4e_5x5"
bottom: "inception_4e_pool_proj"
top: "inception_4e_output"
name: "inception_4e_output"
type: CONCAT
}
layers {
bottom: "inception_4e_output"
top: "pool4"
name: "pool4"
type: POOLING
pooling_param {
pool: MAX
kernel_size: 3
stride: 2
}
}
# ================== hierarchy 5(a) ==================
# ====================================================
# conv 1 x 1
# ----------
layers {
bottom: "pool4"
top: "inception_5a_1x1"
name: "inception_5a_1x1"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 256
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_5a_1x1"
top: "inception_5a_1x1"
name: "inception_5a_relu_1x1"
type: RELU
}
# conv 3 x 3
# ----------
layers {
bottom: "pool4"
top: "inception_5a_3x3_reduce"
name: "inception_5a_3x3_reduce"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 160
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.09
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_5a_3x3_reduce"
top: "inception_5a_3x3_reduce"
name: "inception_5a_relu_3x3_reduce"
type: RELU
}
layers {
bottom: "inception_5a_3x3_reduce"
top: "inception_5a_3x3"
name: "inception_5a_3x3"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 320
pad: 1
kernel_size: 3
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_5a_3x3"
top: "inception_5a_3x3"
name: "inception_5a_relu_3x3"
type: RELU
}
# conv 5 x 5
# ----------
layers {
bottom: "pool4"
top: "inception_5a_5x5_reduce"
name: "inception_5a_5x5_reduce"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 32
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.2
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_5a_5x5_reduce"
top: "inception_5a_5x5_reduce"
name: "inception_5a_relu_5x5_reduce"
type: RELU
}
layers {
bottom: "inception_5a_5x5_reduce"
top: "inception_5a_5x5"
name: "inception_5a_5x5"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 128
pad: 2
kernel_size: 5
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_5a_5x5"
top: "inception_5a_5x5"
name: "inception_5a_relu_5x5"
type: RELU
}
# pool 3 x 3
# ----------
layers {
bottom: "pool4"
top: "inception_5a_pool"
name: "inception_5a_pool"
type: POOLING
pooling_param {
pool: MAX
kernel_size: 3
stride: 1
pad: 1
}
}
layers {
bottom: "inception_5a_pool"
top: "inception_5a_pool_proj"
name: "inception_5a_pool_proj"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 128
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.1
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_5a_pool_proj"
top: "inception_5a_pool_proj"
name: "inception_5a_relu_pool_proj"
type: RELU
}
# concatenate them all
# --------------------
layers {
bottom: "inception_5a_1x1"
bottom: "inception_5a_3x3"
bottom: "inception_5a_5x5"
bottom: "inception_5a_pool_proj"
top: "inception_5a_output"
name: "inception_5a_output"
type: CONCAT
}
# ================== hierarchy 5(b) ==================
# ====================================================
# conv 1 x 1
# ----------
layers {
bottom: "inception_5a_output"
top: "inception_5b_1x1"
name: "inception_5b_1x1"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 384
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_5b_1x1"
top: "inception_5b_1x1"
name: "inception_5b_relu_1x1"
type: RELU
}
# conv 3 x 3
# ----------
layers {
bottom: "inception_5a_output"
top: "inception_5b_3x3_reduce"
name: "inception_5b_3x3_reduce"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 192
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.09
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_5b_3x3_reduce"
top: "inception_5b_3x3_reduce"
name: "inception_5b_relu_3x3_reduce"
type: RELU
}
layers {
bottom: "inception_5b_3x3_reduce"
top: "inception_5b_3x3"
name: "inception_5b_3x3"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 384
pad: 1
kernel_size: 3
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_5b_3x3"
top: "inception_5b_3x3"
name: "inception_5b_relu_3x3"
type: RELU
}
# conv 5 x 5
# ----------
layers {
bottom: "inception_5a_output"
top: "inception_5b_5x5_reduce"
name: "inception_5b_5x5_reduce"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 48
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.2
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_5b_5x5_reduce"
top: "inception_5b_5x5_reduce"
name: "inception_5b_relu_5x5_reduce"
type: RELU
}
layers {
bottom: "inception_5b_5x5_reduce"
top: "inception_5b_5x5"
name: "inception_5b_5x5"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 128
pad: 2
kernel_size: 5
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_5b_5x5"
top: "inception_5b_5x5"
name: "inception_5b_relu_5x5"
type: RELU
}
# pool 3 x 3
# ----------
layers {
bottom: "inception_5a_output"
top: "inception_5b_pool"
name: "inception_5b_pool"
type: POOLING
pooling_param {
pool: MAX
kernel_size: 3
stride: 1
pad: 1
}
}
layers {
bottom: "inception_5b_pool"
top: "inception_5b_pool_proj"
name: "inception_5b_pool_proj"
type: CONVOLUTION
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
convolution_param {
num_output: 128
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.1
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layers {
bottom: "inception_5b_pool_proj"
top: "inception_5b_pool_proj"
name: "inception_5b_relu_pool_proj"
type: RELU
}
# concatenate them all
# --------------------
layers {
bottom: "inception_5b_1x1"
bottom: "inception_5b_3x3"
bottom: "inception_5b_5x5"
bottom: "inception_5b_pool_proj"
top: "inception_5b_output"
name: "inception_5b_output"
type: CONCAT
}
# ===================== softmax 2 ====================
# ====================================================
# ave_pool -> droput -> fc -> loss, accuracy
layers {
bottom: "inception_5b_output"
top: "pool5"
name: "pool5"
type: POOLING
pooling_param {
pool: AVE
kernel_size: 7
stride: 1
}
}
layers {
bottom: "pool5"
top: "pool5"
name: "pool5_drop"
type: DROPOUT
dropout_param {
dropout_ratio: 0.4
}
}
layers {
bottom: "pool5"
top: "loss3_classifier_model"
name: "loss3_classifier_model"
type: INNER_PRODUCT
blobs_lr: 1
blobs_lr: 2
weight_decay: 1
weight_decay: 0
inner_product_param {
num_output: 431
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0
}
}
}
# loss
layers {
bottom: "loss3_classifier_model"
bottom: "label"
top: "loss3"
name: "loss3"
type: SOFTMAX_LOSS
loss_weight: 1
}
# accuracy
layers {
bottom: "loss3_classifier_model"
bottom: "label"
top: "loss3_top-1"
name: "loss3_top-1"
type: ACCURACY
include {
phase: TEST
}
}
layers {
bottom: "loss3_classifier_model"
bottom: "label"
top: "loss3_top-5"
name: "loss3_top-5"
type: ACCURACY
accuracy_param {
top_k: 5
}
include {
phase: TEST
}
}
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment