Skip to content

GitLab

Commit 8c7476d9 authored by Xianzhi Du's avatar Xianzhi Du Committed by A. Unique TensorFlower
Browse files

Internal change 

PiperOrigin-RevId: 374481243
parent 2321ad71
Hide whitespace changes
Inline Side-by-side
Showing with 69 additions and 42 deletions
official/vision/beta/configs/experiments/maskrcnn/coco_spinenet143_casrcnn_tpu.yaml → official/vision/beta/configs/experiments/maskrcnn/coco_spinenet143_cascadercnn_tpu.yaml
View file @ 8c7476d9
# Expect to reach: box mAP: 51.6%, mask mAP: 44.5% on COCO
# --experiment_type=cascadercnn_spinenet_coco
# Expect to reach: box mAP: 51.9%, mask mAP: 45.0% on COCO
runtime:
distribution_strategy: 'tpu'
mixed_precision_dtype: 'bfloat16'
......@@ -43,12 +44,12 @@ task:
detection_generator:
pre_nms_top_k: 1000
trainer:
train_steps: 162050
train_steps: 231000
optimizer_config:
learning_rate:
type: 'stepwise'
stepwise:
boundaries: [148160, 157420]
boundaries: [219450, 226380]
values: [0.32, 0.032, 0.0032]
warmup:
type: 'linear'
......
official/vision/beta/configs/experiments/maskrcnn/coco_spinenet49_casrcnn_tpu.yaml → official/vision/beta/configs/experiments/maskrcnn/coco_spinenet49_cascadercnn_tpu.yaml
View file @ 8c7476d9
# --experiment_type=cascadercnn_spinenet_coco
# Expect to reach: box mAP: 46.4%, mask mAP: 40.0% on COCO
runtime:
distribution_strategy: 'tpu'
mixed_precision_dtype: 'bfloat16'
......@@ -48,7 +50,7 @@ trainer:
type: 'stepwise'
stepwise:
boundaries: [219450, 226380]
values: [0.28, 0.028, 0.0028]
values: [0.32, 0.032, 0.0032]
warmup:
type: 'linear'
linear:
......
official/vision/beta/configs/maskrcnn.py
View file @ 8c7476d9
......@@ -233,6 +233,8 @@ def fasterrcnn_resnetfpn_coco() -> cfg.ExperimentConfig:
"""COCO object detection with Faster R-CNN."""
steps_per_epoch = 500
coco_val_samples = 5000
train_batch_size = 64
eval_batch_size = 8
config = cfg.ExperimentConfig(
runtime=cfg.RuntimeConfig(mixed_precision_dtype='bfloat16'),
......@@ -252,16 +254,16 @@ def fasterrcnn_resnetfpn_coco() -> cfg.ExperimentConfig:
train_data=DataConfig(
input_path=os.path.join(COCO_INPUT_PATH_BASE, 'train*'),
is_training=True,
global_batch_size=64,
global_batch_size=train_batch_size,
parser=Parser(
aug_rand_hflip=True, aug_scale_min=0.8, aug_scale_max=1.25)),
validation_data=DataConfig(
input_path=os.path.join(COCO_INPUT_PATH_BASE, 'val*'),
is_training=False,
global_batch_size=8)),
global_batch_size=eval_batch_size)),
trainer=cfg.TrainerConfig(
train_steps=22500,
validation_steps=coco_val_samples // 8,
validation_steps=coco_val_samples // eval_batch_size,
validation_interval=steps_per_epoch,
steps_per_loop=steps_per_epoch,
summary_interval=steps_per_epoch,
......@@ -300,6 +302,8 @@ def maskrcnn_resnetfpn_coco() -> cfg.ExperimentConfig:
"""COCO object detection with Mask R-CNN."""
steps_per_epoch = 500
coco_val_samples = 5000
train_batch_size = 64
eval_batch_size = 8
config = cfg.ExperimentConfig(
runtime=cfg.RuntimeConfig(mixed_precision_dtype='bfloat16'),
......@@ -314,16 +318,16 @@ def maskrcnn_resnetfpn_coco() -> cfg.ExperimentConfig:
train_data=DataConfig(
input_path=os.path.join(COCO_INPUT_PATH_BASE, 'train*'),
is_training=True,
global_batch_size=64,
global_batch_size=train_batch_size,
parser=Parser(
aug_rand_hflip=True, aug_scale_min=0.8, aug_scale_max=1.25)),
validation_data=DataConfig(
input_path=os.path.join(COCO_INPUT_PATH_BASE, 'val*'),
is_training=False,
global_batch_size=8)),
global_batch_size=eval_batch_size)),
trainer=cfg.TrainerConfig(
train_steps=22500,
validation_steps=coco_val_samples // 8,
validation_steps=coco_val_samples // eval_batch_size,
validation_interval=steps_per_epoch,
steps_per_loop=steps_per_epoch,
summary_interval=steps_per_epoch,
......@@ -357,40 +361,50 @@ def maskrcnn_resnetfpn_coco() -> cfg.ExperimentConfig:
return config
@exp_factory.register_config_factory('cascadercnn_resnetfpn_coco')
def cascadercnn_resnetfpn_coco() -> cfg.ExperimentConfig:
"""COCO object detection with Cascade R-CNN."""
steps_per_epoch = 500
@exp_factory.register_config_factory('maskrcnn_spinenet_coco')
def maskrcnn_spinenet_coco() -> cfg.ExperimentConfig:
"""COCO object detection with Mask R-CNN with SpineNet backbone."""
steps_per_epoch = 463
coco_val_samples = 5000
train_batch_size = 256
eval_batch_size = 8
config = cfg.ExperimentConfig(
runtime=cfg.RuntimeConfig(mixed_precision_dtype='bfloat16'),
task=MaskRCNNTask(
init_checkpoint='gs://cloud-tpu-checkpoints/vision-2.0/resnet50_imagenet/ckpt-28080',
init_checkpoint_modules='backbone',
annotation_file=os.path.join(COCO_INPUT_PATH_BASE,
'instances_val2017.json'),
model=MaskRCNN(
backbone=backbones.Backbone(
type='spinenet',
spinenet=backbones.SpineNet(
model_id='49',
min_level=3,
max_level=7,
)),
decoder=decoders.Decoder(
type='identity', identity=decoders.Identity()),
anchor=Anchor(anchor_size=3),
norm_activation=common.NormActivation(use_sync_bn=True),
num_classes=91,
input_size=[1024, 1024, 3],
include_mask=True,
roi_sampler=ROISampler(cascade_iou_thresholds=[0.6, 0.7]),
detection_head=DetectionHead(
class_agnostic_bbox_pred=True, cascade_class_ensemble=True)),
input_size=[640, 640, 3],
min_level=3,
max_level=7,
include_mask=True),
losses=Losses(l2_weight_decay=0.00004),
train_data=DataConfig(
input_path=os.path.join(COCO_INPUT_PATH_BASE, 'train*'),
is_training=True,
global_batch_size=64,
global_batch_size=train_batch_size,
parser=Parser(
aug_rand_hflip=True, aug_scale_min=0.8, aug_scale_max=1.25)),
aug_rand_hflip=True, aug_scale_min=0.5, aug_scale_max=2.0)),
validation_data=DataConfig(
input_path=os.path.join(COCO_INPUT_PATH_BASE, 'val*'),
is_training=False,
global_batch_size=8)),
global_batch_size=eval_batch_size)),
trainer=cfg.TrainerConfig(
train_steps=22500,
validation_steps=coco_val_samples // 8,
train_steps=steps_per_epoch * 350,
validation_steps=coco_val_samples // eval_batch_size,
validation_interval=steps_per_epoch,
steps_per_loop=steps_per_epoch,
summary_interval=steps_per_epoch,
......@@ -405,30 +419,36 @@ def cascadercnn_resnetfpn_coco() -> cfg.ExperimentConfig:
'learning_rate': {
'type': 'stepwise',
'stepwise': {
'boundaries': [15000, 20000],
'values': [0.12, 0.012, 0.0012],
'boundaries': [
steps_per_epoch * 320, steps_per_epoch * 340
],
'values': [0.32, 0.032, 0.0032],
}
},
'warmup': {
'type': 'linear',
'linear': {
'warmup_steps': 500,
'warmup_steps': 2000,
'warmup_learning_rate': 0.0067
}
}
})),
restrictions=[
'task.train_data.is_training != None',
'task.validation_data.is_training != None'
'task.validation_data.is_training != None',
'task.model.min_level == task.model.backbone.spinenet.min_level',
'task.model.max_level == task.model.backbone.spinenet.max_level',
])
return config
@exp_factory.register_config_factory('maskrcnn_spinenet_coco')
def maskrcnn_spinenet_coco() -> cfg.ExperimentConfig:
"""COCO object detection with Mask R-CNN with SpineNet backbone."""
@exp_factory.register_config_factory('cascadercnn_spinenet_coco')
def cascadercnn_spinenet_coco() -> cfg.ExperimentConfig:
"""COCO object detection with Cascade R-CNN with SpineNet backbone."""
steps_per_epoch = 463
coco_val_samples = 5000
train_batch_size = 256
eval_batch_size = 8
config = cfg.ExperimentConfig(
runtime=cfg.RuntimeConfig(mixed_precision_dtype='bfloat16'),
......@@ -445,8 +465,12 @@ def maskrcnn_spinenet_coco() -> cfg.ExperimentConfig:
)),
decoder=decoders.Decoder(
type='identity', identity=decoders.Identity()),
roi_sampler=ROISampler(cascade_iou_thresholds=[0.6, 0.7]),
detection_head=DetectionHead(
class_agnostic_bbox_pred=True, cascade_class_ensemble=True),
anchor=Anchor(anchor_size=3),
norm_activation=common.NormActivation(use_sync_bn=True),
norm_activation=common.NormActivation(
use_sync_bn=True, activation='swish'),
num_classes=91,
input_size=[640, 640, 3],
min_level=3,
......@@ -456,16 +480,16 @@ def maskrcnn_spinenet_coco() -> cfg.ExperimentConfig:
train_data=DataConfig(
input_path=os.path.join(COCO_INPUT_PATH_BASE, 'train*'),
is_training=True,
global_batch_size=256,
global_batch_size=train_batch_size,
parser=Parser(
aug_rand_hflip=True, aug_scale_min=0.5, aug_scale_max=2.0)),
aug_rand_hflip=True, aug_scale_min=0.1, aug_scale_max=2.5)),
validation_data=DataConfig(
input_path=os.path.join(COCO_INPUT_PATH_BASE, 'val*'),
is_training=False,
global_batch_size=8)),
global_batch_size=eval_batch_size)),
trainer=cfg.TrainerConfig(
train_steps=steps_per_epoch * 350,
validation_steps=coco_val_samples // 8,
train_steps=steps_per_epoch * 500,
validation_steps=coco_val_samples // eval_batch_size,
validation_interval=steps_per_epoch,
steps_per_loop=steps_per_epoch,
summary_interval=steps_per_epoch,
......@@ -481,9 +505,9 @@ def maskrcnn_spinenet_coco() -> cfg.ExperimentConfig:
'type': 'stepwise',
'stepwise': {
'boundaries': [
steps_per_epoch * 320, steps_per_epoch * 340
steps_per_epoch * 475, steps_per_epoch * 490
],
'values': [0.28, 0.028, 0.0028],
'values': [0.32, 0.032, 0.0032],
}
},
'warmup': {
......
official/vision/beta/configs/maskrcnn_test.py
View file @ 8c7476d9
......@@ -29,7 +29,7 @@ class MaskRCNNConfigTest(tf.test.TestCase, parameterized.TestCase):
('fasterrcnn_resnetfpn_coco',),
('maskrcnn_resnetfpn_coco',),
('maskrcnn_spinenet_coco',),
('cascadercnn_resnetfpn_coco',),
('cascadercnn_spinenet_coco',),
)
def test_maskrcnn_configs(self, config_name):
config = exp_factory.get_exp_config(config_name)
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment