def apply_transforms(img_path, pipeline_cfg):
"""Apply transforms pipeline and get both formatted data and the image
without formatting."""
data = dict(img_info=dict(filename=img_path), img_prefix=None)
def split_pipeline_cfg(pipeline_cfg):
"""to split the transfoms into image_transforms and
format_transforms."""
image_transforms_cfg, format_transforms_cfg = [], []
if pipeline_cfg[0]['type'] != 'LoadImageFromFile':
pipeline_cfg.insert(0, dict(type='LoadImageFromFile'))
for transform in pipeline_cfg:
if transform['type'] in FORMAT_TRANSFORMS_SET:
format_transforms_cfg.append(transform)
else:
image_transforms_cfg.append(transform)
return image_transforms_cfg, format_transforms_cfg
image_transforms, format_transforms = split_pipeline_cfg(pipeline_cfg)
image_transforms = Compose(image_transforms)
format_transforms = Compose(format_transforms)
intermediate_data = image_transforms(data)
inference_img = copy.deepcopy(intermediate_data['img'])
format_data = format_transforms(intermediate_data)
return format_data, inference_img
def test_randomgrayscale():
# test rgb2gray, return the grayscale image with p>1
in_img = np.random.rand(10, 10, 3).astype(np.float32)
kwargs = dict(gray_prob=2)
aug = []
aug.extend([mmcls_transforms.RandomGrayscale(**kwargs)])
composed_transform = Compose(aug)
print(composed_transform)
results = dict()
results['img'] = in_img
img = composed_transform(results)['img']
computed_gray = (in_img[:, :, 0] * 0.299 + in_img[:, :, 1] * 0.587 +
in_img[:, :, 2] * 0.114)
for i in range(img.shape[2]):
assert_array_almost_equal(img[:, :, i], computed_gray, decimal=4)
assert img.shape == (10, 10, 3)
# test rgb2gray, return the original image with p=-1
in_img = np.random.rand(10, 10, 3).astype(np.float32)
kwargs = dict(gray_prob=-1)
aug = []
aug.extend([mmcls_transforms.RandomGrayscale(**kwargs)])
composed_transform = Compose(aug)
results = dict()
results['img'] = in_img
img = composed_transform(results)['img']
assert_array_equal(img, in_img)
assert img.shape == (10, 10, 3)
# test image with one channel with our method
# and the function from torchvision
in_img = np.random.rand(10, 10, 1).astype(np.float32)
kwargs = dict(gray_prob=2)
aug = []
aug.extend([mmcls_transforms.RandomGrayscale(**kwargs)])
composed_transform = Compose(aug)
results = dict()
results['img'] = in_img
img = composed_transform(results)['img']
assert_array_equal(img, in_img)
assert img.shape == (10, 10, 1)
in_img_pil = Image.fromarray(in_img[:, :, 0], mode='L')
kwargs = dict(p=2)
aug = []
aug.extend([torchvision.transforms.RandomGrayscale(**kwargs)])
composed_transform = Compose(aug)
img_pil = composed_transform(in_img_pil)
assert_array_equal(np.array(img_pil), np.array(in_img_pil))
assert np.array(img_pil).shape == (10, 10)
def inference_model(model, img):
"""Inference image(s) with the classifier.
Args:
model (nn.Module): The loaded classifier.
img (str/ndarray): The image filename.
Returns:
result (dict): The classification results that contains
`class_name`, `pred_label` and `pred_score`.
"""
cfg = model.cfg
device = next(model.parameters()).device # model device
# build the data pipeline
test_pipeline = Compose(cfg.data.test.pipeline)
# prepare data
data = dict(img_info=dict(filename=img), img_prefix=None)
data = test_pipeline(data)
data = collate([data], samples_per_gpu=1)
if next(model.parameters()).is_cuda:
# scatter to specified GPU
data = scatter(data, [device])[0]
# forward the model
with torch.no_grad():
scores = model(return_loss=False, **data)
pred_score = np.max(scores, axis=1)[0]
pred_label = np.argmax(scores, axis=1)[0]
result = {'pred_label': pred_label, 'pred_score': pred_score}
result['class_name'] = model.CLASSES[result['pred_label']]
return result
def inference_model_batch(model, images_nps, topn=5):
"""Inference image(s) with the classifier.
Args:
model (nn.Module): The loaded classifier.
img (str/ndarray): The image filename or loaded image.
Returns:
result (list of dict): The classification results that contains
`class_name`, `pred_label` and `pred_score`.
"""
cfg = model.cfg
device = next(model.parameters()).device # model device
# build the data pipeline
if cfg.data.test.pipeline[0]['type'] == 'LoadImageFromFile':
cfg.data.test.pipeline.pop(0)
test_pipeline = Compose(cfg.data.test.pipeline)
with torch.no_grad():
inference_results = []
for images_batch in sly.batched(images_nps, g.batch_size):
data = [dict(img=img) for img in images_batch]
data = [test_pipeline(row) for row in data]
data = collate(data, samples_per_gpu=1)
if next(model.parameters()).is_cuda:
# scatter to specified GPU
data = scatter(data, [device])[0]
batch_scores = np.asarray(model(return_loss=False, **data))
batch_top_indexes = batch_scores.argsort(axis=1)[:,
-topn:][:, ::-1]
for scores, top_indexes in zip(batch_scores, batch_top_indexes):
inference_results.append({
'label':
top_indexes.astype(int).tolist(),
'score':
scores[top_indexes].astype(float).tolist(),
'class':
np.asarray(model.CLASSES)[top_indexes].tolist()
})
return inference_results
def build_dataset_pipeline(cfg, phase):
"""build dataset and pipeline from config.
Separate the pipeline except 'LoadImageFromFile' step if
'LoadImageFromFile' in the pipeline.
"""
data_cfg = cfg.data[phase]
loadimage_pipeline = []
if len(data_cfg.pipeline
) != 0 and data_cfg.pipeline[0]['type'] == 'LoadImageFromFile':
loadimage_pipeline.append(data_cfg.pipeline.pop(0))
origin_pipeline = data_cfg.pipeline
data_cfg.pipeline = loadimage_pipeline
dataset = build_dataset(data_cfg)
pipeline = Compose(origin_pipeline)
return dataset, pipeline
def make_img_transform(img_size,
norm_cfg=imagenet_norm,
to_rgb=True,
divisor=32):
img_size = tuple(img_size)
imw, imh = img_size
return Compose([
AlbuAug([
A.LongestMaxSize(max_size=imw),
A.PadIfNeeded(min_height=imh,
min_width=imw,
border_mode=cv2.BORDER_REPLICATE),
]),
Normalize(**norm_cfg, to_rgb=to_rgb),
ImageToTensor(['img']),
Collect(['img']),
])
def inference(config_file, checkpoint, classes, args):
cfg = Config.fromfile(config_file)
model = init_model(cfg, checkpoint, device=args.device)
model.CLASSES = classes
# build the data pipeline
if cfg.data.test.pipeline[0]['type'] != 'LoadImageFromFile':
cfg.data.test.pipeline.insert(0, dict(type='LoadImageFromFile'))
if cfg.data.test.type in ['CIFAR10', 'CIFAR100']:
# The image shape of CIFAR is (32, 32, 3)
cfg.data.test.pipeline.insert(1, dict(type='Resize', size=32))
data = dict(img_info=dict(filename=args.img), img_prefix=None)
test_pipeline = Compose(cfg.data.test.pipeline)
data = test_pipeline(data)
resolution = tuple(data['img'].shape[1:])
data = collate([data], samples_per_gpu=1)
if next(model.parameters()).is_cuda:
# scatter to specified GPU
data = scatter(data, [args.device])[0]
# forward the model
result = {'resolution': resolution}
with torch.no_grad():
if args.inference_time:
time_record = []
for _ in range(10):
start = time()
scores = model(return_loss=False, **data)
time_record.append((time() - start) * 1000)
result['time_mean'] = np.mean(time_record[1:-1])
result['time_std'] = np.std(time_record[1:-1])
else:
scores = model(return_loss=False, **data)
pred_score = np.max(scores, axis=1)[0]
pred_label = np.argmax(scores, axis=1)[0]
result['pred_label'] = pred_label
result['pred_score'] = float(pred_score)
result['pred_class'] = model.CLASSES[result['pred_label']]
result['model'] = config_file.stem
return result
def inference_model(model, img, topn=5):
"""Inference image(s) with the classifier.
Args:
model (nn.Module): The loaded classifier.
img (str/ndarray): The image filename or loaded image.
Returns:
result (list of dict): The classification results that contains
`class_name`, `pred_label` and `pred_score`.
"""
cfg = model.cfg
device = next(model.parameters()).device # model device
# build the data pipeline
if isinstance(img, str):
if cfg.data.test.pipeline[0]['type'] != 'LoadImageFromFile':
cfg.data.test.pipeline.insert(0, dict(type='LoadImageFromFile'))
data = dict(img_info=dict(filename=img), img_prefix=None)
else:
if cfg.data.test.pipeline[0]['type'] == 'LoadImageFromFile':
cfg.data.test.pipeline.pop(0)
data = dict(img=img)
test_pipeline = Compose(cfg.data.test.pipeline)
data = test_pipeline(data)
data = collate([data], samples_per_gpu=1)
if next(model.parameters()).is_cuda:
# scatter to specified GPU
data = scatter(data, [device])[0]
# forward the model
with torch.no_grad():
scores = model(return_loss=False, **data)
model_out = scores[0]
top_scores = model_out[model_out.argsort()[-topn:]][::-1]
top_labels = model_out.argsort()[-topn:][::-1]
result = []
for label, score in zip(top_labels, top_scores):
result.append({
'label': int(label),
'score': float(score),
'class': model.CLASSES[label]
})
return result
def test_randomresizedcrop():
ori_img = mmcv.imread(osp.join(osp.dirname(__file__), '../data/color.jpg'),
'color')
ori_img_pil = Image.open(
osp.join(osp.dirname(__file__), '../data/color.jpg'))
seed = random.randint(0, 100)
# test when scale is not of kind (min, max)
with pytest.raises(ValueError):
kwargs = dict(size=(200, 300),
scale=(1.0, 0.08),
ratio=(3. / 4., 4. / 3.))
aug = []
aug.extend([mmcls_transforms.RandomResizedCrop(**kwargs)])
composed_transform = Compose(aug)
results = dict()
results['img'] = ori_img
composed_transform(results)['img']
# test when ratio is not of kind (min, max)
with pytest.raises(ValueError):
kwargs = dict(size=(200, 300),
scale=(0.08, 1.0),
ratio=(4. / 3., 3. / 4.))
aug = []
aug.extend([mmcls_transforms.RandomResizedCrop(**kwargs)])
composed_transform = Compose(aug)
results = dict()
results['img'] = ori_img
composed_transform(results)['img']
# test crop size is int
kwargs = dict(size=200, scale=(0.08, 1.0), ratio=(3. / 4., 4. / 3.))
random.seed(seed)
np.random.seed(seed)
aug = []
aug.extend([torchvision.transforms.RandomResizedCrop(**kwargs)])
composed_transform = Compose(aug)
baseline = composed_transform(ori_img_pil)
random.seed(seed)
np.random.seed(seed)
aug = []
aug.extend([mmcls_transforms.RandomResizedCrop(**kwargs)])
composed_transform = Compose(aug)
# test __repr__()
print(composed_transform)
results = dict()
results['img'] = ori_img
img = composed_transform(results)['img']
assert np.array(img).shape == (200, 200, 3)
assert np.array(baseline).shape == (200, 200, 3)
nonzero = len((ori_img - np.array(ori_img_pil)[:, :, ::-1]).nonzero())
nonzero_transform = len((img - np.array(baseline)[:, :, ::-1]).nonzero())
assert nonzero == nonzero_transform
# test crop size < image size
kwargs = dict(size=(200, 300), scale=(0.08, 1.0), ratio=(3. / 4., 4. / 3.))
random.seed(seed)
np.random.seed(seed)
aug = []
aug.extend([torchvision.transforms.RandomResizedCrop(**kwargs)])
composed_transform = Compose(aug)
baseline = composed_transform(ori_img_pil)
random.seed(seed)
np.random.seed(seed)
aug = []
aug.extend([mmcls_transforms.RandomResizedCrop(**kwargs)])
composed_transform = Compose(aug)
results = dict()
results['img'] = ori_img
img = composed_transform(results)['img']
assert np.array(img).shape == (200, 300, 3)
assert np.array(baseline).shape == (200, 300, 3)
nonzero = len((ori_img - np.array(ori_img_pil)[:, :, ::-1]).nonzero())
nonzero_transform = len((img - np.array(baseline)[:, :, ::-1]).nonzero())
assert nonzero == nonzero_transform
# test crop size > image size
kwargs = dict(size=(600, 700), scale=(0.08, 1.0), ratio=(3. / 4., 4. / 3.))
random.seed(seed)
np.random.seed(seed)
aug = []
aug.extend([torchvision.transforms.RandomResizedCrop(**kwargs)])
composed_transform = Compose(aug)
baseline = composed_transform(ori_img_pil)
random.seed(seed)
np.random.seed(seed)
aug = []
aug.extend([mmcls_transforms.RandomResizedCrop(**kwargs)])
composed_transform = Compose(aug)
results = dict()
results['img'] = ori_img
img = composed_transform(results)['img']
assert np.array(img).shape == (600, 700, 3)
assert np.array(baseline).shape == (600, 700, 3)
nonzero = len((ori_img - np.array(ori_img_pil)[:, :, ::-1]).nonzero())
nonzero_transform = len((img - np.array(baseline)[:, :, ::-1]).nonzero())
assert nonzero == nonzero_transform
# test cropping the whole image
kwargs = dict(size=(ori_img.shape[0], ori_img.shape[1]),
scale=(1.0, 2.0),
ratio=(1.0, 2.0))
random.seed(seed)
np.random.seed(seed)
aug = []
aug.extend([torchvision.transforms.RandomResizedCrop(**kwargs)])
composed_transform = Compose(aug)
baseline = composed_transform(ori_img_pil)
random.seed(seed)
np.random.seed(seed)
aug = []
aug.extend([mmcls_transforms.RandomResizedCrop(**kwargs)])
composed_transform = Compose(aug)
results = dict()
results['img'] = ori_img
img = composed_transform(results)['img']
assert np.array(img).shape == (ori_img.shape[0], ori_img.shape[1], 3)
assert np.array(baseline).shape == (ori_img.shape[0], ori_img.shape[1], 3)
nonzero = len((ori_img - np.array(ori_img_pil)[:, :, ::-1]).nonzero())
nonzero_transform = len((img - np.array(baseline)[:, :, ::-1]).nonzero())
assert nonzero == nonzero_transform
# assert_array_equal(ori_img, img)
# assert_array_equal(np.array(ori_img_pil), np.array(baseline))
# test central crop when in_ratio < min(ratio)
kwargs = dict(size=(ori_img.shape[0], ori_img.shape[1]),
scale=(1.0, 2.0),
ratio=(2, 3))
random.seed(seed)
np.random.seed(seed)
aug = []
aug.extend([torchvision.transforms.RandomResizedCrop(**kwargs)])
composed_transform = Compose(aug)
baseline = composed_transform(ori_img_pil)
random.seed(seed)
np.random.seed(seed)
aug = []
aug.extend([mmcls_transforms.RandomResizedCrop(**kwargs)])
composed_transform = Compose(aug)
results = dict()
results['img'] = ori_img
img = composed_transform(results)['img']
assert np.array(img).shape == (ori_img.shape[0], ori_img.shape[1], 3)
assert np.array(baseline).shape == (ori_img.shape[0], ori_img.shape[1], 3)
nonzero = len((ori_img - np.array(ori_img_pil)[:, :, ::-1]).nonzero())
nonzero_transform = len((img - np.array(baseline)[:, :, ::-1]).nonzero())
assert nonzero == nonzero_transform
# test central crop when in_ratio > max(ratio)
kwargs = dict(size=(ori_img.shape[0], ori_img.shape[1]),
scale=(1.0, 2.0),
ratio=(3. / 4, 1))
random.seed(seed)
np.random.seed(seed)
aug = []
aug.extend([torchvision.transforms.RandomResizedCrop(**kwargs)])
composed_transform = Compose(aug)
baseline = composed_transform(ori_img_pil)
random.seed(seed)
np.random.seed(seed)
aug = []
aug.extend([mmcls_transforms.RandomResizedCrop(**kwargs)])
composed_transform = Compose(aug)
results = dict()
results['img'] = ori_img
img = composed_transform(results)['img']
assert np.array(img).shape == (ori_img.shape[0], ori_img.shape[1], 3)
assert np.array(baseline).shape == (ori_img.shape[0], ori_img.shape[1], 3)
nonzero = len((ori_img - np.array(ori_img_pil)[:, :, ::-1]).nonzero())
nonzero_transform = len((img - np.array(baseline)[:, :, ::-1]).nonzero())
assert nonzero == nonzero_transform
# test different interpolation types
for mode in ['nearest', 'bilinear', 'bicubic', 'area', 'lanczos']:
kwargs = dict(size=(600, 700),
scale=(0.08, 1.0),
ratio=(3. / 4., 4. / 3.),
interpolation=mode)
aug = []
aug.extend([mmcls_transforms.RandomResizedCrop(**kwargs)])
composed_transform = Compose(aug)
results = dict()
results['img'] = ori_img
img = composed_transform(results)['img']
assert img.shape == (600, 700, 3)
def test_randomcrop():
ori_img = mmcv.imread(osp.join(osp.dirname(__file__), '../data/color.jpg'),
'color')
ori_img_pil = Image.open(
osp.join(osp.dirname(__file__), '../data/color.jpg'))
seed = random.randint(0, 100)
# test crop size is int
kwargs = dict(size=200, padding=0, pad_if_needed=True, fill=0)
random.seed(seed)
np.random.seed(seed)
aug = []
aug.extend([torchvision.transforms.RandomCrop(**kwargs)])
composed_transform = Compose(aug)
baseline = composed_transform(ori_img_pil)
kwargs = dict(size=200, padding=0, pad_if_needed=True, pad_val=0)
random.seed(seed)
np.random.seed(seed)
aug = []
aug.extend([mmcls_transforms.RandomCrop(**kwargs)])
composed_transform = Compose(aug)
# test __repr__()
print(composed_transform)
results = dict()
results['img'] = ori_img
img = composed_transform(results)['img']
assert np.array(img).shape == (200, 200, 3)
assert np.array(baseline).shape == (200, 200, 3)
nonzero = len((ori_img - np.array(ori_img_pil)[:, :, ::-1]).nonzero())
nonzero_transform = len((img - np.array(baseline)[:, :, ::-1]).nonzero())
assert nonzero == nonzero_transform
# test crop size < image size
kwargs = dict(size=(200, 300), padding=0, pad_if_needed=True, fill=0)
random.seed(seed)
np.random.seed(seed)
aug = []
aug.extend([torchvision.transforms.RandomCrop(**kwargs)])
composed_transform = Compose(aug)
baseline = composed_transform(ori_img_pil)
kwargs = dict(size=(200, 300), padding=0, pad_if_needed=True, pad_val=0)
random.seed(seed)
np.random.seed(seed)
aug = []
aug.extend([mmcls_transforms.RandomCrop(**kwargs)])
composed_transform = Compose(aug)
results = dict()
results['img'] = ori_img
img = composed_transform(results)['img']
assert np.array(img).shape == (200, 300, 3)
assert np.array(baseline).shape == (200, 300, 3)
nonzero = len((ori_img - np.array(ori_img_pil)[:, :, ::-1]).nonzero())
nonzero_transform = len((img - np.array(baseline)[:, :, ::-1]).nonzero())
assert nonzero == nonzero_transform
# test crop size > image size
kwargs = dict(size=(600, 700), padding=0, pad_if_needed=True, fill=0)
random.seed(seed)
np.random.seed(seed)
aug = []
aug.extend([torchvision.transforms.RandomCrop(**kwargs)])
composed_transform = Compose(aug)
baseline = composed_transform(ori_img_pil)
kwargs = dict(size=(600, 700), padding=0, pad_if_needed=True, pad_val=0)
random.seed(seed)
np.random.seed(seed)
aug = []
aug.extend([mmcls_transforms.RandomCrop(**kwargs)])
composed_transform = Compose(aug)
results = dict()
results['img'] = ori_img
img = composed_transform(results)['img']
assert np.array(img).shape == (600, 700, 3)
assert np.array(baseline).shape == (600, 700, 3)
nonzero = len((ori_img - np.array(ori_img_pil)[:, :, ::-1]).nonzero())
nonzero_transform = len((img - np.array(baseline)[:, :, ::-1]).nonzero())
assert nonzero == nonzero_transform
# test crop size == image size
kwargs = dict(size=(ori_img.shape[0], ori_img.shape[1]),
padding=0,
pad_if_needed=True,
fill=0)
random.seed(seed)
np.random.seed(seed)
aug = []
aug.extend([torchvision.transforms.RandomCrop(**kwargs)])
composed_transform = Compose(aug)
baseline = composed_transform(ori_img_pil)
kwargs = dict(size=(ori_img.shape[0], ori_img.shape[1]),
padding=0,
pad_if_needed=True,
pad_val=0)
random.seed(seed)
np.random.seed(seed)
aug = []
aug.extend([mmcls_transforms.RandomCrop(**kwargs)])
composed_transform = Compose(aug)
results = dict()
results['img'] = ori_img
img = composed_transform(results)['img']
assert np.array(img).shape == (img.shape[0], img.shape[1], 3)
assert np.array(baseline).shape == (img.shape[0], img.shape[1], 3)
nonzero = len((ori_img - np.array(ori_img_pil)[:, :, ::-1]).nonzero())
nonzero_transform = len((img - np.array(baseline)[:, :, ::-1]).nonzero())
assert nonzero == nonzero_transform
assert_array_equal(ori_img, img)
assert_array_equal(np.array(baseline), np.array(ori_img_pil))
# test different padding mode
for mode in ['constant', 'edge', 'reflect', 'symmetric']:
kwargs = dict(size=(500, 600), padding=0, pad_if_needed=True, fill=0)
kwargs['padding_mode'] = mode
random.seed(seed)
np.random.seed(seed)
aug = []
aug.extend([torchvision.transforms.RandomCrop(**kwargs)])
composed_transform = Compose(aug)
baseline = composed_transform(ori_img_pil)
kwargs = dict(size=(500, 600),
padding=0,
pad_if_needed=True,
pad_val=0)
random.seed(seed)
np.random.seed(seed)
aug = []
aug.extend([mmcls_transforms.RandomCrop(**kwargs)])
composed_transform = Compose(aug)
results = dict()
results['img'] = ori_img
img = composed_transform(results)['img']
assert np.array(img).shape == (500, 600, 3)
assert np.array(baseline).shape == (500, 600, 3)
nonzero = len((ori_img - np.array(ori_img_pil)[:, :, ::-1]).nonzero())
nonzero_transform = len(
(img - np.array(baseline)[:, :, ::-1]).nonzero())
assert nonzero == nonzero_transform