Skip to content
GitLab
Menu
Projects
Groups
Snippets
Loading...
Help
Help
Support
Community forum
Keyboard shortcuts
?
Submit feedback
Contribute to GitLab
Sign in / Register
Toggle navigation
Menu
Open sidebar
ModelZoo
ResNet50_tensorflow
Commits
d68c7b96
"...resnet50_tensorflow.git" did not exist on "e73a2d0293ce4f40437d18f03f8186d682d4822f"
Commit
d68c7b96
authored
Oct 06, 2020
by
Shixin Luo
Browse files
add default config for mobilenet v2 that replicate the expected performance
parent
7a9c72ad
Changes
2
Hide whitespace changes
Inline
Side-by-side
Showing
2 changed files
with
62 additions
and
11 deletions
+62
-11
official/vision/beta/configs/experiments/image_classification/imagenet_mobilenetv2_1.0_gpu.yaml
...ts/image_classification/imagenet_mobilenetv2_1.0_gpu.yaml
+10
-11
official/vision/beta/configs/experiments/image_classification/imagenet_mobilenetv2_1.0_tpu.yaml
...ts/image_classification/imagenet_mobilenetv2_1.0_tpu.yaml
+52
-0
No files found.
official/vision/beta/configs/experiments/image_classification/imagenet_mobilenetv2_1.0_gpu.yaml
View file @
d68c7b96
# MobileNetV2_1.0 ImageNet classification. 71.0% top-1 and 90.0% top-5 accuracy.
runtime
:
distribution_strategy
:
'
mirrored'
mixed_precision_dtype
:
'
float
32
'
mixed_precision_dtype
:
'
float
16
'
loss_scale
:
'
dynamic'
task
:
model
:
...
...
@@ -20,15 +21,15 @@ task:
input_path
:
'
imagenet-2012-tfrecord/train*'
is_training
:
True
global_batch_size
:
768
# 96 * 8
dtype
:
'
float
32
'
dtype
:
'
float
16
'
validation_data
:
input_path
:
'
imagenet-2012-tfrecord/valid*'
is_training
:
False
global_batch_size
:
768
# 96 * 8
dtype
:
'
float
32
'
dtype
:
'
float
16
'
drop_remainder
:
False
trainer
:
train_steps
:
7
00000
# 700K
train_steps
:
3
00000
validation_steps
:
65
validation_interval
:
1668
steps_per_loop
:
1668
# NUM_EXAMPLES (1281167) // global_batch_size
...
...
@@ -36,19 +37,17 @@ trainer:
checkpoint_interval
:
1668
optimizer_config
:
optimizer
:
type
:
'
rmsprop'
rmsprop
:
decay
:
0.9
type
:
'
sgd'
sgd
:
momentum
:
0.9
epsilon
:
0.002
learning_rate
:
type
:
'
exponential'
exponential
:
initial_learning_rate
:
0.
36
# 0.
045
* NUM_GPUS
initial_learning_rate
:
0.
8
# 0.
1
* NUM_GPUS
decay_steps
:
4170
# 2.5 * steps_per_epoch
decay_rate
:
0.9
8
decay_rate
:
0.9
5
staircase
:
True
warmup
:
type
:
'
linear'
linear
:
warmup_steps
:
8340
# 5 * steps_per_epoch
\ No newline at end of file
warmup_steps
:
100
\ No newline at end of file
official/vision/beta/configs/experiments/image_classification/imagenet_mobilenetv2_1.0_tpu.yaml
0 → 100644
View file @
d68c7b96
# MobileNetV2_1.0 ImageNet classification. 72.26% top-1 and 90.76% top-5 accuracy.
runtime
:
distribution_strategy
:
'
tpu'
mixed_precision_dtype
:
'
bfloat16'
task
:
model
:
num_classes
:
1001
input_size
:
[
224
,
224
,
3
]
backbone
:
type
:
'
mobilenet'
mobilenet
:
model_id
:
'
MobileNetV2'
width_multiplier
:
1.0
dropout_rate
:
0.2
losses
:
l2_weight_decay
:
0.00002
one_hot
:
True
label_smoothing
:
0.1
train_data
:
input_path
:
'
gs://tf_mobilenet/imagenet/imagenet-2012-tfrecord/train*'
is_training
:
True
global_batch_size
:
2048
# 64 * 32
dtype
:
'
bfloat16'
validation_data
:
input_path
:
'
gs://tf_mobilenet/imagenet/imagenet-2012-tfrecord/valid*'
is_training
:
False
global_batch_size
:
2048
# 64 * 32
dtype
:
'
bfloat16'
drop_remainder
:
False
trainer
:
train_steps
:
300000
validation_steps
:
24
validation_interval
:
625
steps_per_loop
:
625
# NUM_EXAMPLES (1281167) // global_batch_size
summary_interval
:
625
checkpoint_interval
:
625
optimizer_config
:
optimizer
:
type
:
'
sgd'
sgd
:
momentum
:
0.9
learning_rate
:
type
:
'
exponential'
exponential
:
initial_learning_rate
:
3.2
# 0.1 * 32
decay_steps
:
1562
# 2.5 * steps_per_epoch
decay_rate
:
0.96
staircase
:
True
warmup
:
type
:
'
linear'
linear
:
warmup_steps
:
50
\ No newline at end of file
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
.
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment