def _load_file(self, f):
# pdb.set_trace()
# catalog lookup
if f.startswith("catalog://"):
paths_catalog = import_file(
"maskrcnn_benchmark.config.paths_catalog", self.cfg.PATHS_CATALOG, True
)
catalog_f = paths_catalog.ModelCatalog.get(f[len("catalog://") :])
self.logger.info("{} points to {}".format(f, catalog_f))
f = catalog_f
# download url files
if f.startswith("http"):
# if the file is a url path, download it and cache it
cached_f = cache_url(f)
self.logger.info("url {} cached in {}".format(f, cached_f))
f = cached_f
# convert Caffe2 checkpoint from pkl
if f.endswith(".pkl"):
return load_c2_format(self.cfg, f)
# pdb.set_trace()
# load native detectron.pytorch checkpoint
loaded = super(DetectronCheckpointer, self)._load_file(f)
# pdb.set_trace()
if "model" not in loaded:
loaded = dict(model=loaded)
# pdb.set_trace()
return loaded
def _load_file(self, f):
# catalog lookup
if f.startswith("catalog://"):
from maskrcnn_benchmark.config import paths_catalog
catalog_f = paths_catalog.ModelCatalog.get(f[len("catalog://"):])
self.logger.info("{} points to {}".format(f, catalog_f))
f = catalog_f
# download url files
if f.startswith("http"):
# if the file is a url path, download it and cache it
cached_f = cache_url(f)
self.logger.info("url {} cached in {}".format(f, cached_f))
f = cached_f
# convert Caffe2 checkpoint from pkl
if f.endswith(".pkl"):
return load_c2_format(self.cfg, f)
# load native detectron.pytorch checkpoint
loaded = super(DetectronCheckpointer, self)._load_file(f)
if "model" not in loaded:
loaded = dict(model=loaded)
if self.cfg is not None:
from .c2_model_loading import _rename_conv_weights_for_deformable_conv_layers
_rename_conv_weights_for_deformable_conv_layers(
loaded['model'], self.cfg)
return loaded
def load_checkpoint(self, weights, is_restore=False):
t_start = time.time()
if weights.endswith(".pkl"):
# for caffe2 model
from maskrcnn_benchmark.utils.c2_model_loading import \
load_c2_format
loaded = load_c2_format(self.cfg, weights)
else:
loaded = torch.load(weights, map_location=torch.device("cpu"))
t_io_end = time.time()
if "model" not in loaded:
loaded = dict(model=loaded)
self.state.model = load_model(self.state.model,
loaded['model'],
self.logger,
is_restore=is_restore)
if "optimizer" in loaded:
self.state.optimizer.load_state_dict(loaded['optimizer'])
if "iteration" in loaded:
self.state.iteration = loaded['iteration']
if "scheduler" in loaded:
self.state.scheduler.load_state_dict(loaded["scheduler"])
del loaded
t_end = time.time()
self.logger.info("Load checkpoint from file {}, "
"Time usage:\n\tIO: {}, restore snapshot: {}".format(
weights, t_io_end - t_start, t_end - t_io_end))
def _load_file(self, f):
# catalog lookup
if f.startswith("catalog://"):
paths_catalog = import_file(
"maskrcnn_benchmark.config.paths_catalog",
self.cfg.PATHS_CATALOG, True)
catalog_f = paths_catalog.ModelCatalog.get(f[len("catalog://"):])
self.logger.info("{} points to {}".format(f, catalog_f))
f = catalog_f
# download url files
if f.startswith("http"):
# if the file is a url path, download it and cache it
cached_f = cache_url(f)
self.logger.info("url {} cached in {}".format(f, cached_f))
f = cached_f
# convert Caffe2 checkpoint from pkl
if f.endswith(".pkl"):
print
return load_c2_format(self.cfg, f)
# load native detectron.pytorch checkpoint
loaded = super(DetectronCheckpointer, self)._load_file(f)
# unload_keyword = list(self.cfg.FEW_SHOT.UNLOAD_KEYWORD)
# if self.cfg.FEW_SHOT.LOAD_PRETRIANED_RPN_ONLY:
# unload_keyword += ['roi_heads']
# new_dict = {}
# for key,value in loaded.items():
# if any([ keyword in key for keyword in unload_keyword]):
# continue
# new_dict[key] = value
# loaded = new_dict
if "model" not in loaded:
loaded = dict(model=loaded)
return loaded
def _load_file(self, f):
# f: catalog://ImageNetPretrained/MSRA/R-50
if f.startswith("catalog://"):
# 把paths_catalog.py作为一个module导入, 这个文件的路径是配置文件中设置的
# 有可能存在于文件系统的任何位置, 因此不能直接import
paths_catalog = import_file(
"maskrcnn_benchmark.config.paths_catalog",
self.cfg.PATHS_CATALOG, # paths_catalog.py的绝对路径
True)
# 'https://dl.fbaipublicfiles.com/detectron/ImageNetPretrained/MSRA/R-50.pkl'
# 获取pretrained model的url
catalog_f = paths_catalog.ModelCatalog.get(f[len("catalog://"):])
self.logger.info("{} points to {}".format(f, catalog_f))
f = catalog_f
# download url files
if f.startswith("http"):
# if the file is a url path, download it and cache it
cached_f = cache_url(f)
self.logger.info("url {} cached in {}".format(f, cached_f))
f = cached_f
# 把pkl文件转换成Caffe2模型文件
if f.endswith(".pkl"):
return load_c2_format(self.cfg, f)
# load native detectron.pytorch checkpoint
loaded = super(DetectronCheckpointer, self)._load_file(f)
if "model" not in loaded:
loaded = dict(model=loaded)
return loaded
def _load_file(cfg):
f = cfg.MODEL.WEIGHT
# catalog lookup
if f.startswith("catalog://"):
paths_catalog = import_file("maskrcnn_benchmark.config.paths_catalog",
cfg.PATHS_CATALOG, True)
catalog_f = paths_catalog.ModelCatalog.get(f[len("catalog://"):])
f = catalog_f
# download url files
if f.startswith("http"):
# if the file is a url path, download it and cache it
cached_f = cache_url(f)
f = cached_f
# convert Caffe2 checkpoint from pkl
if f.endswith(".pkl"):
return load_c2_format(cfg, f)
def _load_file(self, f):
# catalog lookup
if f.startswith("catalog://"):
paths_catalog = import_file(
"maskrcnn_benchmark.config.paths_catalog", self.cfg.PATHS_CATALOG, True
)
catalog_f = paths_catalog.ModelCatalog.get(f[len("catalog://") :])
self.logger.info("{} points to {}".format(f, catalog_f))
f = catalog_f
# download url files
if f.startswith("http"):
# if the file is a url path, download it and cache it
cached_f = cache_url(f)
self.logger.info("url {} cached in {}".format(f, cached_f))
f = cached_f
# convert Caffe2 checkpoint from pkl
if f.endswith(".pkl"):
return load_c2_format(self.cfg, f)
# load native detectron.pytorch checkpoint
loaded = super(DetectronCheckpointer, self)._load_file(f)
if "model" not in loaded:
loaded = dict(model=loaded)
return loaded
import torch as t from maskrcnn_benchmark.config import cfg from maskrcnn_benchmark.utils.c2_model_loading import load_c2_format path = "FAST_RCNN_MLP_DIM2048_FPN_DIM512.pkl" config_name = 'modified_for_conversion_e2e_faster_rcnn_X-101-64x4d-FPN_1x_MLP_2048_FPN_512.yaml' base_path = '/private/home/meetshah/detectron/vmb/configs/visual_genome_vqa/c2/' cfg.merge_from_file(base_path + config_name) _d = load_c2_format(cfg, path) newdict = _d t.save(newdict, "model_final.pth")
default="~/.torch/models/_detectron_35858933_12_2017_baselines_e2e_mask_rcnn_R-50-FPN_1x.yaml.01_48_14.DzEQe4wC_output_train_coco_2014_train%3Acoco_2014_valminusminival_generalized_rcnn_model_final.pkl",
help="path to detectron pretrained weight(.pkl)",
type=str,
)
parser.add_argument(
"--save_path",
default="./pretrained_model/no_last_layers.pth",
help="path to save the converted model",
type=str,
)
parser.add_argument(
"--cfg",
default="../ configs/e2e_mask_rcnn_R_50_FPN_1x.yaml",
help="path to config file",
type=str,
)
args = parser.parse_args()
#
DETECTRON_PATH = os.path.expanduser(args.pretrained_path)
print('detectron path: {}'.format(DETECTRON_PATH))
cfg.merge_from_file(args.cfg)
_d = load_c2_format(cfg, DETECTRON_PATH)
newdict = _d
newdict['model'] = removekey(_d['model'],
['cls_score.bias', 'cls_score.weight', 'bbox_pred.bias', 'bbox_pred.weight'])
torch.save(newdict, args.save_path)
print('saved to {}.'.format(args.save_path))
