def build(image_set, args):
root = Path(args.coco_path)
assert root.exists(), f'provided COCO path {root} does not exist'
mode = 'instances'
PATHS = {
"train": (root / "train", root / "annotations" / f'{mode}_train.json'),
"val": (root / "val", root / "annotations" / f'{mode}_val.json'),
}
img_folder, ann_file = PATHS[image_set]
dataset = CocoDetection(img_folder, ann_file, transforms=make_coco_transforms(image_set), return_masks=args.masks,
cache_mode=args.cache_mode, local_rank=get_local_rank(), local_size=get_local_size())
return dataset
def build(args, img_folder, ann_file, image_set, activated_class_ids,
with_support):
return DetectionDataset(args,
img_folder,
ann_file,
transforms=make_transforms(image_set),
support_transforms=make_support_transforms(),
return_masks=False,
activated_class_ids=activated_class_ids,
with_support=with_support,
cache_mode=args.cache_mode,
local_rank=get_local_rank(),
local_size=get_local_size())
def build(image_set, args):
root = Path("/home/eini/WY/project/Deformable-DETR/data/MOT15")
# assert root.exists(), f'provided COCO path {root} does not exist'
# mode = 'instances'
PATHS = {
"train": (root / "images/train", root / "labels_with_ids" / 'train.json'),
"val": (root / "images/val", root / "labels_with_ids" / 'val.json'),
}
img_folder, ann_file = PATHS[image_set]
dataset = MOT15Detection(img_folder, ann_file, transforms=make_mot15_transforms(image_set), return_masks=args.masks,
cache_mode=args.cache_mode, local_rank=get_local_rank(), local_size=get_local_size())
return dataset
def build_support_dataset(image_set, args):
if not args.fewshot_finetune:
assert image_set == "train"
if args.dataset_file == 'coco':
root = Path('data/coco/')
img_folder = root / "train2017"
ann_file = root / "annotations" / "instances_train2017.json"
return SupportDataset(img_folder, ann_file,
activatedClassIds=coco_base_class_ids+coco_novel_class_ids,
transforms=make_support_transforms(),
cache_mode=args.cache_mode,
local_rank=get_local_rank(),
local_size=get_local_size())
if args.dataset_file == 'coco_base':
root = Path('data/coco/')
img_folder = root / "train2017"
ann_file = root / "annotations" / "instances_train2017.json"
return SupportDataset(img_folder, ann_file,
activatedClassIds=coco_base_class_ids,
transforms=make_support_transforms(),
cache_mode=args.cache_mode,
local_rank=get_local_rank(),
local_size=get_local_size())
if args.dataset_file == 'voc':
root = Path('data/voc')
img_folder = root / "images"
ann_files = [root / "annotations" / 'pascal_train2007.json',
root / "annotations" / 'pascal_val2007.json',
root / "annotations" / 'pascal_train2012.json',
root / "annotations" / 'pascal_val2012.json']
return SupportDataset(img_folder, ann_files,
activatedClassIds=list(range(1, 20+1)),
transforms=make_support_transforms(),
cache_mode=args.cache_mode,
local_rank=get_local_rank(),
local_size=get_local_size())
if args.dataset_file == 'voc_base1':
root = Path('data/voc')
img_folder = root / "images"
ann_files = [root / "annotations" / 'pascal_train2007.json',
root / "annotations" / 'pascal_val2007.json',
root / "annotations" / 'pascal_train2012.json',
root / "annotations" / 'pascal_val2012.json']
return SupportDataset(img_folder, ann_files,
activatedClassIds=voc_base1_class_ids,
transforms=make_support_transforms(),
cache_mode=args.cache_mode,
local_rank=get_local_rank(),
local_size=get_local_size())
if args.dataset_file == 'voc_base2':
root = Path('data/voc')
img_folder = root / "images"
ann_files = [root / "annotations" / 'pascal_train2007.json',
root / "annotations" / 'pascal_val2007.json',
root / "annotations" / 'pascal_train2012.json',
root / "annotations" / 'pascal_val2012.json']
return SupportDataset(img_folder, ann_files,
activatedClassIds=voc_base2_class_ids,
transforms=make_support_transforms(),
cache_mode=args.cache_mode,
local_rank=get_local_rank(),
local_size=get_local_size())
if args.dataset_file == 'voc_base3':
root = Path('data/voc')
img_folder = root / "images"
ann_files = [root / "annotations" / 'pascal_train2007.json',
root / "annotations" / 'pascal_val2007.json',
root / "annotations" / 'pascal_train2012.json',
root / "annotations" / 'pascal_val2012.json']
return SupportDataset(img_folder, ann_files,
activatedClassIds=voc_base3_class_ids,
transforms=make_support_transforms(),
cache_mode=args.cache_mode,
local_rank=get_local_rank(),
local_size=get_local_size())
else:
# After Fewshot Fine-tuning, we use the support dataset that was used for few-shot fine-tuning as the support
# dataset for inference (to generate category codes).
assert image_set == "fewshot"
if args.dataset_file == 'coco_base':
root = Path('data/coco_fewshot')
img_folder = root.parent / 'coco' / "train2017"
ids = (coco_base_class_ids + coco_novel_class_ids)
ids.sort()
ann_file = root / f'seed{args.fewshot_seed}' / f'{args.num_shots}shot.json'
return SupportDataset(img_folder, str(ann_file),
activatedClassIds=ids,
transforms=make_support_transforms(),
cache_mode=args.cache_mode,
local_rank=get_local_rank(),
local_size=get_local_size())
if args.dataset_file == 'voc_base1':
root = Path('data/voc_fewshot_split1')
img_folder = root.parent / 'voc' / "images"
ids = list(range(1, 20+1))
ann_file = root / f'seed{args.fewshot_seed}' / f'{args.num_shots}shot.json'
return SupportDataset(img_folder, str(ann_file),
activatedClassIds=ids,
transforms=make_support_transforms(),
cache_mode=args.cache_mode,
local_rank=get_local_rank(),
local_size=get_local_size())
if args.dataset_file == 'voc_base2':
root = Path('data/voc_fewshot_split2')
img_folder = root.parent / 'voc' / "images"
ids = list(range(1, 20+1))
ann_file = root / f'seed{args.fewshot_seed}' / f'{args.num_shots}shot.json'
return SupportDataset(img_folder, str(ann_file),
activatedClassIds=ids,
transforms=make_support_transforms(),
cache_mode=args.cache_mode,
local_rank=get_local_rank(),
local_size=get_local_size())
if args.dataset_file == 'voc_base3':
root = Path('data/voc_fewshot_split3')
img_folder = root.parent / 'voc' / "images"
ids = list(range(1, 20+1))
ann_file = root / f'seed{args.fewshot_seed}' / f'{args.num_shots}shot.json'
return SupportDataset(img_folder, str(ann_file),
activatedClassIds=ids,
transforms=make_support_transforms(),
cache_mode=args.cache_mode,
local_rank=get_local_rank(),
local_size=get_local_size())
raise ValueError
def build(args, image_set, activated_class_ids, with_support=True):
assert image_set == "fewshot"
activated_class_ids.sort()
if args.dataset_file in ['coco_base']:
root = Path('data/coco_fewshot')
img_folder = root.parent / 'coco' / "train2017"
ann_file = root / f'seed{args.fewshot_seed}' / f'{args.num_shots}shot.json'
return DetectionDataset(args,
img_folder,
str(ann_file),
transforms=make_transforms(),
support_transforms=make_support_transforms(),
return_masks=False,
activated_class_ids=activated_class_ids,
with_support=with_support,
cache_mode=args.cache_mode,
local_rank=get_local_rank(),
local_size=get_local_size())
if args.dataset_file == "voc_base1":
root = Path('data/voc_fewshot_split1')
img_folder = root.parent / 'voc' / "images"
ann_file = root / f'seed{args.fewshot_seed}' / f'{args.num_shots}shot.json'
return DetectionDataset(args,
img_folder,
str(ann_file),
transforms=make_transforms(),
support_transforms=make_support_transforms(),
return_masks=False,
activated_class_ids=activated_class_ids,
with_support=with_support,
cache_mode=args.cache_mode,
local_rank=get_local_rank(),
local_size=get_local_size())
if args.dataset_file == "voc_base2":
root = Path('data/voc_fewshot_split2')
img_folder = root.parent / 'voc' / "images"
ann_file = root / f'seed{args.fewshot_seed}' / f'{args.num_shots}shot.json'
return DetectionDataset(args,
img_folder,
str(ann_file),
transforms=make_transforms(),
support_transforms=make_support_transforms(),
return_masks=False,
activated_class_ids=activated_class_ids,
with_support=with_support,
cache_mode=args.cache_mode,
local_rank=get_local_rank(),
local_size=get_local_size())
if args.dataset_file == "voc_base3":
root = Path('data/voc_fewshot_split3')
img_folder = root.parent / 'voc' / "images"
ann_file = root / f'seed{args.fewshot_seed}' / f'{args.num_shots}shot.json'
return DetectionDataset(args,
img_folder,
str(ann_file),
transforms=make_transforms(),
support_transforms=make_support_transforms(),
return_masks=False,
activated_class_ids=activated_class_ids,
with_support=with_support,
cache_mode=args.cache_mode,
local_rank=get_local_rank(),
local_size=get_local_size())
raise ValueError