Skip to content

GitLab

".github/git@developer.sourcefind.cn:change/sglang.git" did not exist on "4ad97370452e9de7a0f78b246f7d12d7bd2b7d83"
Commit f7938e6a authored by Xianzhi Du's avatar Xianzhi Du Committed by A. Unique TensorFlower
Browse files

Internal change 

PiperOrigin-RevId: 374481243
parent 0d62382b
Show 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 @ f7938e6a
# 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: runtime:
distribution_strategy: 'tpu' distribution_strategy: 'tpu'
mixed_precision_dtype: 'bfloat16' mixed_precision_dtype: 'bfloat16'
...@@ -43,12 +44,12 @@ task: ...@@ -43,12 +44,12 @@ task:
detection_generator: detection_generator:
pre_nms_top_k: 1000 pre_nms_top_k: 1000
trainer: trainer:
train_steps: 162050 train_steps: 231000
optimizer_config: optimizer_config:
learning_rate: learning_rate:
type: 'stepwise' type: 'stepwise'
stepwise: stepwise:
boundaries: [148160, 157420] boundaries: [219450, 226380]
values: [0.32, 0.032, 0.0032] values: [0.32, 0.032, 0.0032]
warmup: warmup:
type: 'linear' 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 @ f7938e6a
# --experiment_type=cascadercnn_spinenet_coco
# Expect to reach: box mAP: 46.4%, mask mAP: 40.0% on COCO
runtime: runtime:
distribution_strategy: 'tpu' distribution_strategy: 'tpu'
mixed_precision_dtype: 'bfloat16' mixed_precision_dtype: 'bfloat16'
...@@ -48,7 +50,7 @@ trainer: ...@@ -48,7 +50,7 @@ trainer:
type: 'stepwise' type: 'stepwise'
stepwise: stepwise:
boundaries: [219450, 226380] boundaries: [219450, 226380]
values: [0.28, 0.028, 0.0028] values: [0.32, 0.032, 0.0032]
warmup: warmup:
type: 'linear' type: 'linear'
linear: linear:
......
official/vision/beta/configs/maskrcnn.py
View file @ f7938e6a
...@@ -233,6 +233,8 @@ def fasterrcnn_resnetfpn_coco() -> cfg.ExperimentConfig: ...@@ -233,6 +233,8 @@ def fasterrcnn_resnetfpn_coco() -> cfg.ExperimentConfig:
"""COCO object detection with Faster R-CNN.""" """COCO object detection with Faster R-CNN."""
steps_per_epoch = 500 steps_per_epoch = 500
coco_val_samples = 5000 coco_val_samples = 5000
train_batch_size = 64
eval_batch_size = 8
config = cfg.ExperimentConfig( config = cfg.ExperimentConfig(
runtime=cfg.RuntimeConfig(mixed_precision_dtype='bfloat16'), runtime=cfg.RuntimeConfig(mixed_precision_dtype='bfloat16'),
...@@ -252,16 +254,16 @@ def fasterrcnn_resnetfpn_coco() -> cfg.ExperimentConfig: ...@@ -252,16 +254,16 @@ def fasterrcnn_resnetfpn_coco() -> cfg.ExperimentConfig:
train_data=DataConfig( train_data=DataConfig(
input_path=os.path.join(COCO_INPUT_PATH_BASE, 'train*'), input_path=os.path.join(COCO_INPUT_PATH_BASE, 'train*'),
is_training=True, is_training=True,
global_batch_size=64, global_batch_size=train_batch_size,
parser=Parser( parser=Parser(
aug_rand_hflip=True, aug_scale_min=0.8, aug_scale_max=1.25)), aug_rand_hflip=True, aug_scale_min=0.8, aug_scale_max=1.25)),
validation_data=DataConfig( validation_data=DataConfig(
input_path=os.path.join(COCO_INPUT_PATH_BASE, 'val*'), input_path=os.path.join(COCO_INPUT_PATH_BASE, 'val*'),
is_training=False, is_training=False,
global_batch_size=8)), global_batch_size=eval_batch_size)),
trainer=cfg.TrainerConfig( trainer=cfg.TrainerConfig(
train_steps=22500, train_steps=22500,
validation_steps=coco_val_samples // 8, validation_steps=coco_val_samples // eval_batch_size,
validation_interval=steps_per_epoch, validation_interval=steps_per_epoch,
steps_per_loop=steps_per_epoch, steps_per_loop=steps_per_epoch,
summary_interval=steps_per_epoch, summary_interval=steps_per_epoch,
...@@ -300,6 +302,8 @@ def maskrcnn_resnetfpn_coco() -> cfg.ExperimentConfig: ...@@ -300,6 +302,8 @@ def maskrcnn_resnetfpn_coco() -> cfg.ExperimentConfig:
"""COCO object detection with Mask R-CNN.""" """COCO object detection with Mask R-CNN."""
steps_per_epoch = 500 steps_per_epoch = 500
coco_val_samples = 5000 coco_val_samples = 5000
train_batch_size = 64
eval_batch_size = 8
config = cfg.ExperimentConfig( config = cfg.ExperimentConfig(
runtime=cfg.RuntimeConfig(mixed_precision_dtype='bfloat16'), runtime=cfg.RuntimeConfig(mixed_precision_dtype='bfloat16'),
...@@ -314,16 +318,16 @@ def maskrcnn_resnetfpn_coco() -> cfg.ExperimentConfig: ...@@ -314,16 +318,16 @@ def maskrcnn_resnetfpn_coco() -> cfg.ExperimentConfig:
train_data=DataConfig( train_data=DataConfig(
input_path=os.path.join(COCO_INPUT_PATH_BASE, 'train*'), input_path=os.path.join(COCO_INPUT_PATH_BASE, 'train*'),
is_training=True, is_training=True,
global_batch_size=64, global_batch_size=train_batch_size,
parser=Parser( parser=Parser(
aug_rand_hflip=True, aug_scale_min=0.8, aug_scale_max=1.25)), aug_rand_hflip=True, aug_scale_min=0.8, aug_scale_max=1.25)),
validation_data=DataConfig( validation_data=DataConfig(
input_path=os.path.join(COCO_INPUT_PATH_BASE, 'val*'), input_path=os.path.join(COCO_INPUT_PATH_BASE, 'val*'),
is_training=False, is_training=False,
global_batch_size=8)), global_batch_size=eval_batch_size)),
trainer=cfg.TrainerConfig( trainer=cfg.TrainerConfig(
train_steps=22500, train_steps=22500,
validation_steps=coco_val_samples // 8, validation_steps=coco_val_samples // eval_batch_size,
validation_interval=steps_per_epoch, validation_interval=steps_per_epoch,
steps_per_loop=steps_per_epoch, steps_per_loop=steps_per_epoch,
summary_interval=steps_per_epoch, summary_interval=steps_per_epoch,
...@@ -357,40 +361,50 @@ def maskrcnn_resnetfpn_coco() -> cfg.ExperimentConfig: ...@@ -357,40 +361,50 @@ def maskrcnn_resnetfpn_coco() -> cfg.ExperimentConfig:
return config return config
@exp_factory.register_config_factory('cascadercnn_resnetfpn_coco') @exp_factory.register_config_factory('maskrcnn_spinenet_coco')
def cascadercnn_resnetfpn_coco() -> cfg.ExperimentConfig: def maskrcnn_spinenet_coco() -> cfg.ExperimentConfig:
"""COCO object detection with Cascade R-CNN.""" """COCO object detection with Mask R-CNN with SpineNet backbone."""
steps_per_epoch = 500 steps_per_epoch = 463
coco_val_samples = 5000 coco_val_samples = 5000
train_batch_size = 256
eval_batch_size = 8
config = cfg.ExperimentConfig( config = cfg.ExperimentConfig(
runtime=cfg.RuntimeConfig(mixed_precision_dtype='bfloat16'), runtime=cfg.RuntimeConfig(mixed_precision_dtype='bfloat16'),
task=MaskRCNNTask( 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, annotation_file=os.path.join(COCO_INPUT_PATH_BASE,
'instances_val2017.json'), 'instances_val2017.json'),
model=MaskRCNN( 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, num_classes=91,
input_size=[1024, 1024, 3], input_size=[640, 640, 3],
include_mask=True, min_level=3,
roi_sampler=ROISampler(cascade_iou_thresholds=[0.6, 0.7]), max_level=7,
detection_head=DetectionHead( include_mask=True),
class_agnostic_bbox_pred=True, cascade_class_ensemble=True)),
losses=Losses(l2_weight_decay=0.00004), losses=Losses(l2_weight_decay=0.00004),
train_data=DataConfig( train_data=DataConfig(
input_path=os.path.join(COCO_INPUT_PATH_BASE, 'train*'), input_path=os.path.join(COCO_INPUT_PATH_BASE, 'train*'),
is_training=True, is_training=True,
global_batch_size=64, global_batch_size=train_batch_size,
parser=Parser( 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( validation_data=DataConfig(
input_path=os.path.join(COCO_INPUT_PATH_BASE, 'val*'), input_path=os.path.join(COCO_INPUT_PATH_BASE, 'val*'),
is_training=False, is_training=False,
global_batch_size=8)), global_batch_size=eval_batch_size)),
trainer=cfg.TrainerConfig( trainer=cfg.TrainerConfig(
train_steps=22500, train_steps=steps_per_epoch * 350,
validation_steps=coco_val_samples // 8, validation_steps=coco_val_samples // eval_batch_size,
validation_interval=steps_per_epoch, validation_interval=steps_per_epoch,
steps_per_loop=steps_per_epoch, steps_per_loop=steps_per_epoch,
summary_interval=steps_per_epoch, summary_interval=steps_per_epoch,
...@@ -405,30 +419,36 @@ def cascadercnn_resnetfpn_coco() -> cfg.ExperimentConfig: ...@@ -405,30 +419,36 @@ def cascadercnn_resnetfpn_coco() -> cfg.ExperimentConfig:
'learning_rate': { 'learning_rate': {
'type': 'stepwise', 'type': 'stepwise',
'stepwise': { 'stepwise': {
'boundaries': [15000, 20000], 'boundaries': [
'values': [0.12, 0.012, 0.0012], steps_per_epoch * 320, steps_per_epoch * 340
],
'values': [0.32, 0.032, 0.0032],
} }
}, },
'warmup': { 'warmup': {
'type': 'linear', 'type': 'linear',
'linear': { 'linear': {
'warmup_steps': 500, 'warmup_steps': 2000,
'warmup_learning_rate': 0.0067 'warmup_learning_rate': 0.0067
} }
} }
})), })),
restrictions=[ restrictions=[
'task.train_data.is_training != None', '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 return config
@exp_factory.register_config_factory('maskrcnn_spinenet_coco') @exp_factory.register_config_factory('cascadercnn_spinenet_coco')
def maskrcnn_spinenet_coco() -> cfg.ExperimentConfig: def cascadercnn_spinenet_coco() -> cfg.ExperimentConfig:
"""COCO object detection with Mask R-CNN with SpineNet backbone.""" """COCO object detection with Cascade R-CNN with SpineNet backbone."""
steps_per_epoch = 463 steps_per_epoch = 463
coco_val_samples = 5000 coco_val_samples = 5000
train_batch_size = 256
eval_batch_size = 8
config = cfg.ExperimentConfig( config = cfg.ExperimentConfig(
runtime=cfg.RuntimeConfig(mixed_precision_dtype='bfloat16'), runtime=cfg.RuntimeConfig(mixed_precision_dtype='bfloat16'),
...@@ -445,8 +465,12 @@ def maskrcnn_spinenet_coco() -> cfg.ExperimentConfig: ...@@ -445,8 +465,12 @@ def maskrcnn_spinenet_coco() -> cfg.ExperimentConfig:
)), )),
decoder=decoders.Decoder( decoder=decoders.Decoder(
type='identity', identity=decoders.Identity()), 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), 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, num_classes=91,
input_size=[640, 640, 3], input_size=[640, 640, 3],
min_level=3, min_level=3,
...@@ -456,16 +480,16 @@ def maskrcnn_spinenet_coco() -> cfg.ExperimentConfig: ...@@ -456,16 +480,16 @@ def maskrcnn_spinenet_coco() -> cfg.ExperimentConfig:
train_data=DataConfig( train_data=DataConfig(
input_path=os.path.join(COCO_INPUT_PATH_BASE, 'train*'), input_path=os.path.join(COCO_INPUT_PATH_BASE, 'train*'),
is_training=True, is_training=True,
global_batch_size=256, global_batch_size=train_batch_size,
parser=Parser( 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( validation_data=DataConfig(
input_path=os.path.join(COCO_INPUT_PATH_BASE, 'val*'), input_path=os.path.join(COCO_INPUT_PATH_BASE, 'val*'),
is_training=False, is_training=False,
global_batch_size=8)), global_batch_size=eval_batch_size)),
trainer=cfg.TrainerConfig( trainer=cfg.TrainerConfig(
train_steps=steps_per_epoch * 350, train_steps=steps_per_epoch * 500,
validation_steps=coco_val_samples // 8, validation_steps=coco_val_samples // eval_batch_size,
validation_interval=steps_per_epoch, validation_interval=steps_per_epoch,
steps_per_loop=steps_per_epoch, steps_per_loop=steps_per_epoch,
summary_interval=steps_per_epoch, summary_interval=steps_per_epoch,
...@@ -481,9 +505,9 @@ def maskrcnn_spinenet_coco() -> cfg.ExperimentConfig: ...@@ -481,9 +505,9 @@ def maskrcnn_spinenet_coco() -> cfg.ExperimentConfig:
'type': 'stepwise', 'type': 'stepwise',
'stepwise': { 'stepwise': {
'boundaries': [ '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': { 'warmup': {
......
official/vision/beta/configs/maskrcnn_test.py
View file @ f7938e6a
...@@ -29,7 +29,7 @@ class MaskRCNNConfigTest(tf.test.TestCase, parameterized.TestCase): ...@@ -29,7 +29,7 @@ class MaskRCNNConfigTest(tf.test.TestCase, parameterized.TestCase):
('fasterrcnn_resnetfpn_coco',), ('fasterrcnn_resnetfpn_coco',),
('maskrcnn_resnetfpn_coco',), ('maskrcnn_resnetfpn_coco',),
('maskrcnn_spinenet_coco',), ('maskrcnn_spinenet_coco',),
('cascadercnn_resnetfpn_coco',), ('cascadercnn_spinenet_coco',),
) )
def test_maskrcnn_configs(self, config_name): def test_maskrcnn_configs(self, config_name):
config = exp_factory.get_exp_config(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