def get_samples(cls, roots, ids_dataset=None):
train_samples = []
val_samples = []
test_samples = []
for split in ["train", "val", "test"]:
if isinstance(roots[split]["cap"], tuple):
coco_dataset = (
pyCOCO(roots[split]["cap"][0]),
pyCOCO(roots[split]["cap"][1]),
)
root = roots[split]["img"]
else:
coco_dataset = (pyCOCO(roots[split]["cap"]), )
root = (roots[split]["img"], )
if ids_dataset is None:
ids = list(coco_dataset.anns.keys())
else:
ids = ids_dataset[split]
if isinstance(ids, tuple):
bp = len(ids[0])
ids = list(ids[0]) + list(ids[1])
else:
bp = len(ids)
for index in range(len(ids)):
if index < bp:
coco = coco_dataset[0]
img_root = root[0]
else:
coco = coco_dataset[1]
img_root = root[1]
ann_id = ids[index]
caption = coco.anns[ann_id]["caption"]
img_id = coco.anns[ann_id]["image_id"]
filename = coco.loadImgs(img_id)[0]["file_name"]
example = Example.fromdict({
"image":
os.path.join(img_root, filename),
"text":
caption
})
if split == "train":
train_samples.append(example)
elif split == "val":
val_samples.append(example)
elif split == "test":
test_samples.append(example)
return train_samples, val_samples, test_samples
def get_samples(cls, roots, ids_dataset=None):
train_samples = []
val_samples = []
test_samples = []
for split in ['train', 'val', 'test']:
if isinstance(roots[split]['cap'], tuple):
coco_dataset = (pyCOCO(roots[split]['cap'][0]),
pyCOCO(roots[split]['cap'][1]))
root = roots[split]['img']
else:
coco_dataset = (pyCOCO(roots[split]['cap']), )
root = (roots[split]['img'], )
if ids_dataset is None:
ids = list(coco_dataset.anns.keys())
else:
ids = ids_dataset[split]
if isinstance(ids, tuple):
bp = len(ids[0])
ids = list(ids[0]) + list(ids[1])
else:
bp = len(ids)
for index in range(len(ids)):
if index < bp:
coco = coco_dataset[0]
img_root = root[0]
else:
coco = coco_dataset[1]
img_root = root[1]
ann_id = ids[index]
caption = coco.anns[ann_id]['caption']
img_id = coco.anns[ann_id]['image_id']
filename = coco.loadImgs(img_id)[0]['file_name']
example = Example.fromdict({
'image':
os.path.join(img_root, filename),
'text':
caption
})
if split == 'train':
train_samples.append(example)
elif split == 'val':
val_samples.append(example)
elif split == 'test':
test_samples.append(example)
return train_samples, val_samples, test_samples
def get_samples(cls, roots, imgids_dataset, imgid2captions):
train_samples = []
val_samples = []
test_samples = []
for split in ['train', 'val', 'test']:
coco = pyCOCO(roots[split]['cap'])
img_root = roots[split]['img']
ids = list(coco.anns.keys())
img_ids = imgids_dataset[split]
imgid2count = {i: 0 for i in img_ids}
for index in range(len(ids)):
ann_id = ids[index]
img_id = coco.anns[ann_id]['image_id']
if img_id in img_ids and imgid2count[img_id] < 5:
caption = imgid2captions[str(img_id)][imgid2count[
img_id]] # one caption of that image (preprocessed)
imgid2count[img_id] += 1
filename = coco.loadImgs(img_id)[0]['file_name']
example = Example.fromdict({
'image':
os.path.join(img_root, filename),
'text':
caption
})
if split == 'train':
train_samples.append(example)
elif split == 'val':
val_samples.append(example)
elif split == 'test':
test_samples.append(example)
return train_samples, val_samples, test_samples