def test_pit_model(self):
cfg = CN()
cfg.MODEL = CN()
add_pit_backbone_config(cfg)
build_model = BACKBONE_REGISTRY.get("pit_d2go_model_wrapper")
pit_models = {
"pit_ti_ours": 160,
"pit_ti": 224,
"pit_s_ours_v1": 256,
"pit_s": 224,
}
pit_model_weights = {
"pit_ti_ours":
"manifold://mobile_vision_workflows/tree/workflows/kyungminkim/20210515/deit_[model]pit_scalable_distilled_[bs]128_[mcfg]pit_ti_ours_.HImkjNCpJI/checkpoint_best.pth",
"pit_ti":
"manifold://mobile_vision_workflows/tree/workflows/kyungminkim/20210515/deit_[model]pit_scalable_distilled_[bs]128_[mcfg]pit_ti_.QJeFNUfYOD/checkpoint_best.pth",
"pit_s_ours_v1":
"manifold://mobile_vision_workflows/tree/workflows/kyungminkim/20210515/deit_[model]pit_scalable_distilled_[bs]64_[mcfg]pit_s_ours_v1_.LXdwyBDaNY/checkpoint_best.pth",
"pit_s":
"manifold://mobile_vision_workflows/tree/workflows/kyungminkim/20210515/deit_[model]pit_scalable_distilled_[bs]128_[mcfg]pit_s_.zReQLPOuJe/checkpoint_best.pth",
}
for model_name, org_size in pit_models.items():
print("model_name", model_name)
cfg.MODEL.PIT.MODEL_CONFIG = f"manifold://mobile_vision_workflows/tree/workflows/wbc/deit/model_cfgs/{model_name}.json"
cfg.MODEL.PIT.WEIGHTS = pit_model_weights[model_name]
cfg.MODEL.PIT.DILATED = True
model = build_model(cfg, None)
model.eval()
for input_size_h in [org_size, 192, 224, 256, 320]:
for input_size_w in [org_size, 192, 224, 256, 320]:
x = torch.rand(1, 3, input_size_h, input_size_w)
y = model(x)
print(f"x.shape: {x.shape}, y.shape: {y.shape}")
def cast_from_other_class(cls, other_cfg):
"""Cast an instance of other CfgNode to D2Go's CfgNode (or its subclass)"""
new_cfg = CfgNode(other_cfg)
# copy all fields inside __dict__, this will preserve fields like __deprecated_keys__
for k, v in other_cfg.__dict__.items():
new_cfg.__dict__[k] = v
return new_cfg
def test_diff_cfg_no_new_allowed(self):
"""check that if new_allowed is False, new keys cause key error"""
# create base config
cfg1 = CfgNode()
cfg1.A = CfgNode()
cfg1.A.set_new_allowed(False)
cfg1.A.Y = 2
# case 2: new allowed not set, new config has new keys
cfg2 = cfg1.clone()
cfg2.A.X = 2
self.assertRaises(KeyError, get_diff_cfg, cfg1, cfg2)
def _add_rcnn_default_config(_C):
_C.EXPORT_CAFFE2 = CfgNode()
_C.EXPORT_CAFFE2.USE_HEATMAP_MAX_KEYPOINT = False
# Options about how to export the model
_C.RCNN_EXPORT = CfgNode()
# whether or not to include the postprocess (GeneralizedRCNN._postprocess) step
# inside the exported model
_C.RCNN_EXPORT.INCLUDE_POSTPROCESS = False
_C.RCNN_PREPARE_FOR_EXPORT = "default_rcnn_prepare_for_export"
_C.RCNN_PREPARE_FOR_QUANT = "default_rcnn_prepare_for_quant"
_C.RCNN_PREPARE_FOR_QUANT_CONVERT = "default_rcnn_prepare_for_quant_convert"
def test_modeling_hook_cfg(self):
"""Create model with modeling hook using build_model"""
cfg = CfgNode()
cfg.MODEL = CfgNode()
cfg.MODEL.DEVICE = "cpu"
cfg.MODEL.META_ARCHITECTURE = "TestArch"
cfg.MODEL.MODELING_HOOKS = ["PlusOneHook", "TimesTwoHook"]
model = build_model(cfg)
self.assertEqual(model(2), 10)
self.assertTrue(hasattr(model, "_modeling_hooks"))
self.assertTrue(hasattr(model, "unapply_modeling_hooks"))
orig_model = model.unapply_modeling_hooks()
self.assertIsInstance(orig_model, TestArch)
self.assertEqual(orig_model(2), 4)
def maybe_override_output_dir(cfg: CfgNode, output_dir: Optional[str]) -> None:
"""Overrides the output directory if `output_dir` is not None. """
if output_dir is not None and output_dir != cfg.OUTPUT_DIR:
cfg.OUTPUT_DIR = output_dir
logger.warning(
f"Override cfg.OUTPUT_DIR ({cfg.OUTPUT_DIR}) to be the same as "
f"output_dir {output_dir}")
def maybe_override_output_dir(cfg: CfgNode, output_dir: str):
if cfg.OUTPUT_DIR != output_dir:
with temp_defrost(cfg):
logger.warning(
"Override cfg.OUTPUT_DIR ({}) to be the same as output_dir {}".
format(cfg.OUTPUT_DIR, output_dir))
cfg.OUTPUT_DIR = output_dir
def create_cfg_from_cli_args(args, default_cfg):
"""
Instead of loading from defaults.py, this binary only includes necessary
configs building from scratch, and overrides them from args. There're two
levels of config:
_C: the config system used by this binary, which is a sub-set of training
config, override by configurable_cfg. It can also be override by
args.opts for convinience.
configurable_cfg: common configs that user should explicitly specify
in the args.
"""
_C = CN()
_C.INPUT = default_cfg.INPUT
_C.DATASETS = default_cfg.DATASETS
_C.DATALOADER = default_cfg.DATALOADER
_C.TEST = default_cfg.TEST
if hasattr(default_cfg, "D2GO_DATA"):
_C.D2GO_DATA = default_cfg.D2GO_DATA
if hasattr(default_cfg, "TENSORBOARD"):
_C.TENSORBOARD = default_cfg.TENSORBOARD
# NOTE configs below might not be necessary, but must add to make code work
_C.MODEL = CN()
_C.MODEL.META_ARCHITECTURE = default_cfg.MODEL.META_ARCHITECTURE
_C.MODEL.MASK_ON = default_cfg.MODEL.MASK_ON
_C.MODEL.KEYPOINT_ON = default_cfg.MODEL.KEYPOINT_ON
_C.MODEL.LOAD_PROPOSALS = default_cfg.MODEL.LOAD_PROPOSALS
assert _C.MODEL.LOAD_PROPOSALS is False, "caffe2 model doesn't support"
_C.OUTPUT_DIR = args.output_dir
configurable_cfg = [
"DATASETS.TEST",
args.datasets,
"INPUT.MIN_SIZE_TEST",
args.min_size,
"INPUT.MAX_SIZE_TEST",
args.max_size,
]
cfg = _C.clone()
cfg.merge_from_list(configurable_cfg)
cfg.merge_from_list(args.opts)
return cfg
def run_with_cmdline_args(args):
cfg, output_dir, runner = prepare_for_launch(args)
inference_config = None
if args.inference_config_file:
inference_config = CfgNode(
CfgNode.load_yaml_with_base(args.inference_config_file))
return main(
cfg,
output_dir,
runner,
# binary specific optional arguments
predictor_types=args.predictor_types,
compare_accuracy=args.compare_accuracy,
skip_if_fail=args.skip_if_fail,
inference_config=inference_config,
)
def test_diff_cfg_with_new_allowed(self):
"""diff config with new keys and new_allowed set to True"""
# create base config
cfg1 = CfgNode()
cfg1.A = CfgNode()
cfg1.A.set_new_allowed(True)
cfg1.A.Y = 2
# case 3: new allowed set, new config has new keys
cfg2 = cfg1.clone()
cfg2.A.X = 2
gt = CfgNode()
gt.A = CfgNode()
gt.A.X = 2
self.assertEqual(gt, get_diff_cfg(cfg1, cfg2))
def test_get_diff_cfg(self):
"""check config that is diff from default config, no new keys"""
# create base config
cfg1 = CfgNode()
cfg1.A = CfgNode()
cfg1.A.Y = 2
# case 1: new allowed not set, new config has only old keys
cfg2 = cfg1.clone()
cfg2.set_new_allowed(False)
cfg2.A.Y = 3
gt = CfgNode()
gt.A = CfgNode()
gt.A.Y = 3
self.assertEqual(gt, get_diff_cfg(cfg1, cfg2))
def test_modeling_hook_copy(self):
"""Create model with modeling hook, the model could be copied"""
cfg = CfgNode()
cfg.MODEL = CfgNode()
cfg.MODEL.DEVICE = "cpu"
cfg.MODEL.META_ARCHITECTURE = "TestArch"
cfg.MODEL.MODELING_HOOKS = ["PlusOneHook", "TimesTwoHook"]
model = build_model(cfg)
self.assertEqual(model(2), 10)
model_copy = copy.deepcopy(model)
orig_model = model.unapply_modeling_hooks()
self.assertIsInstance(orig_model, TestArch)
self.assertEqual(orig_model(2), 4)
orig_model_copy = model_copy.unapply_modeling_hooks()
self.assertEqual(orig_model_copy(2), 4)
def setup_after_launch(
cfg: CfgNode,
output_dir: str,
runner: Optional[BaseRunner] = None,
_scale_world_size:
bool = True, # HACK: temporarily allow lightning_train_net to by pass this.
):
"""
Binary-level setup after entering DDP, including
- creating working directory
- setting up logger
- logging environment
- printing and dumping config
- (optional) initializing runner
"""
create_dir_on_global_main_process(output_dir)
setup_loggers(output_dir)
log_system_info()
cfg.freeze()
maybe_override_output_dir(cfg, output_dir)
logger.info("Running with full config:\n{}".format(cfg))
dump_cfg(cfg, os.path.join(output_dir, "config.yaml"))
if runner:
logger.info("Initializing runner ...")
runner = initialize_runner(runner, cfg)
logger.info("Running with runner: {}".format(runner))
# save the diff config
if runner:
default_cfg = runner.get_default_cfg()
dump_cfg(
get_diff_cfg(default_cfg, cfg),
os.path.join(output_dir, "diff_config.yaml"),
)
else:
# TODO: support getting default_cfg without runner.
pass
# scale the config after dumping so that dumped config files keep original world size
if _scale_world_size:
auto_scale_world_size(cfg, new_world_size=comm.get_world_size())
def test_merge_from_list_with_new_allowed(self):
"""
YACS's merge_from_list doesn't take new_allowed into account, D2Go override its behavior, and this test covers it.
"""
# new_allowed is not set
cfg = CfgNode()
cfg.A = CfgNode()
cfg.A.X = 1
self.assertRaises(Exception, cfg.merge_from_list, ["A.Y", "2"])
# new_allowed is set for sub key
cfg = CfgNode()
cfg.A = CfgNode(new_allowed=True)
cfg.A.X = 1
cfg.merge_from_list(["A.Y", "2"])
self.assertEqual(cfg.A.Y,
2) # note that the string will be converted to number
# however new_allowed is not set for root key
self.assertRaises(Exception, cfg.merge_from_list, ["B", "3"])
def get_default_cfg():
_C = super(GeneralizedRCNNRunner, GeneralizedRCNNRunner).get_default_cfg()
_C.EXPORT_CAFFE2 = CfgNode()
_C.EXPORT_CAFFE2.USE_HEATMAP_MAX_KEYPOINT = False
_C.RCNN_PREPARE_FOR_EXPORT = "default_rcnn_prepare_for_export"
_C.RCNN_PREPARE_FOR_QUANT = "default_rcnn_prepare_for_quant"
_C.RCNN_PREPARE_FOR_QUANT_CONVERT = "default_rcnn_prepare_for_quant_convert"
return _C
def _setup_after_launch(cfg: CN, output_dir: str, runner):
"""
Set things up after entering DDP, including
- creating working directory
- setting up logger
- logging environment
- initializing runner
"""
create_dir_on_global_main_process(output_dir)
comm.synchronize()
setup_loggers(output_dir)
cfg.freeze()
if cfg.OUTPUT_DIR != output_dir:
with temp_defrost(cfg):
logger.warning(
"Override cfg.OUTPUT_DIR ({}) to be the same as output_dir {}".format(
cfg.OUTPUT_DIR, output_dir
)
)
cfg.OUTPUT_DIR = output_dir
dump_cfg(cfg, os.path.join(output_dir, "config.yaml"))
def get_default_cfg():
"""
Override `get_default_cfg` for adding non common config.
"""
from detectron2.config import get_cfg as get_d2_cfg
cfg = get_d2_cfg()
cfg = CfgNode.cast_from_other_class(
cfg) # upgrade from D2's CfgNode to D2Go's CfgNode
cfg.SOLVER.AUTO_SCALING_METHODS = ["default_scale_d2_configs"]
return cfg
def test_deit_model(self):
cfg = CN()
cfg.MODEL = CN()
add_deit_backbone_config(cfg)
build_model = BACKBONE_REGISTRY.get("deit_d2go_model_wrapper")
deit_models = {
"8X-7-RM_4": 170,
"DeiT-Tiny": 224,
"DeiT-Small": 224,
"32X-1-RM_2": 221,
"8X-7": 160,
"32X-1": 256,
}
deit_model_weights = {
"8X-7-RM_4":
"manifold://mobile_vision_workflows/tree/workflows/kyungminkim/20210511/deit_[model]deit_scaling_distill_[bs]128_[mcfg]8X-7-RM_4_.OIXarYpbZw/checkpoint_best.pth",
"DeiT-Tiny":
"manifold://mobile_vision_workflows/tree/workflows/cl114/DeiT-official-ckpt/deit_tiny_distilled_patch16_224-b40b3cf7.pth",
"DeiT-Small":
"manifold://mobile_vision_workflows/tree/workflows/cl114/DeiT-official-ckpt/deit_small_distilled_patch16_224-649709d9.pth",
"32X-1-RM_2":
"manifold://mobile_vision_workflows/tree/workflows/kyungminkim/20210511/deit_[model]deit_scaling_distill_[bs]64_[mcfg]32X-1-RM_2_.xusuFyNMdD/checkpoint_best.pth",
"8X-7":
"manifold://mobile_vision_workflows/tree/workflows/cl114/scaled_best/8X-7.pth",
"32X-1":
"manifold://mobile_vision_workflows/tree/workflows/cl114/scaled_best/32X-1.pth",
}
for model_name, org_size in deit_models.items():
print("model_name", model_name)
cfg.MODEL.DEIT.MODEL_CONFIG = f"manifold://mobile_vision_workflows/tree/workflows/wbc/deit/model_cfgs/{model_name}.json"
cfg.MODEL.DEIT.WEIGHTS = deit_model_weights[model_name]
model = build_model(cfg, None)
model.eval()
for input_size_h in [org_size, 192, 224, 256, 320]:
for input_size_w in [org_size, 192, 224, 256, 320]:
x = torch.rand(1, 3, input_size_h, input_size_w)
y = model(x)
print(f"x.shape: {x.shape}, y.shape: {y.shape}")
def get_default_cfg():
"""
Override `get_default_cfg` for adding non common config.
"""
from detectron2.config import get_cfg as get_d2_cfg
cfg = get_d2_cfg()
cfg = CfgNode.cast_from_other_class(
cfg) # upgrade from D2's CfgNode to D2Go's CfgNode
try:
from d2go.runner import get_unintentional_added_configs_during_runner_import
for key in get_unintentional_added_configs_during_runner_import():
cfg.register_deprecated_key(key)
except ImportError:
pass
cfg.SOLVER.AUTO_SCALING_METHODS = ["default_scale_d2_configs"]
return cfg
def do_train(cfg: CfgNode, trainer: pl.Trainer,
task: GeneralizedRCNNTask) -> Dict[str, str]:
"""Runs the training loop with given trainer and task.
Args:
cfg: The normalized ConfigNode for this D2Go Task.
trainer: PyTorch Lightning trainer.
task: Lightning module instance.
Returns:
A map of model name to trained model config path.
"""
with EventStorage() as storage:
task.storage = storage
trainer.fit(task)
final_ckpt = os.path.join(cfg.OUTPUT_DIR, FINAL_MODEL_CKPT)
trainer.save_checkpoint(final_ckpt) # for validation monitor
trained_cfg = cfg.clone()
with temp_defrost(trained_cfg):
trained_cfg.MODEL.WEIGHTS = final_ckpt
model_configs = dump_trained_model_configs(
cfg.OUTPUT_DIR, {"model_final": trained_cfg})
return model_configs
def _get_default_config():
cfg = CfgNode()
cfg.INPUT = CfgNode()
cfg.INPUT.CROP = CfgNode()
cfg.INPUT.CROP.ENABLED = False
cfg.INPUT.CROP.SIZE = (0.9, 0.9)
cfg.INPUT.CROP.TYPE = "relative_range"
cfg.MODEL = CfgNode()
cfg.MODEL.MIN_DIM_SIZE = 360
cfg.INFERENCE_SDK = CfgNode()
cfg.INFERENCE_SDK.MODEL = CfgNode()
cfg.INFERENCE_SDK.MODEL.SCORE_THRESHOLD = 0.8
cfg.INFERENCE_SDK.IOU_TRACKER = CfgNode()
cfg.INFERENCE_SDK.IOU_TRACKER.IOU_THRESHOLD = 0.15
cfg.INFERENCE_SDK.ENABLE_ID_TRACKING = True
return cfg
def dump_cfg(cfg: CfgNode, path: str) -> None:
if comm.is_main_process():
with PathManager.open(path, "w") as f:
f.write(cfg.dump())
logger.info("Full config saved to {}".format(path))
def get_default_config():
cfg = CfgNode()
cfg.D2GO_DATA = CfgNode()
cfg.D2GO_DATA.AUG_OPS = CfgNode()
return cfg
def _test1(cfg):
cfg.TEST1 = CfgNode()
cfg.TEST1.X = 1
return cfg
def _test2(cfg):
cfg.TEST2 = CfgNode()
cfg.TEST2.Y = 2
return cfg
