def make_detmot_transforms(image_set, args=None): normalize = T.MotCompose([ T.MotToTensor(), T.MotNormalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]) ]) scales = [608, 640, 672, 704, 736, 768, 800, 832, 864, 896, 928, 960, 992] if image_set == 'train': color_transforms = [] if args.cj: print('Training with RandomColorJitter.') color_transforms.append( T.MoTColorJitter(brightness=0.5, contrast=0.5, saturation=0.5, hue=0)) if not args.crop: scale_transforms = [ T.MotRandomHorizontalFlip(), T.FixedMotRandomShift(bs=1), T.MotRandomResize(scales, max_size=1536), normalize, ] else: print('Training with RandomCrop.') scale_transforms = [ T.MotRandomHorizontalFlip(), T.FixedMotRandomShift(bs=1), T.MotRandomSelect( T.MotRandomResize(scales, max_size=1536), T.MotCompose([ T.MotRandomResize([400, 500, 600]), T.FixedMotRandomCrop(384, 600), T.MotRandomResize(scales, max_size=1536), ])), normalize, ] return T.MotCompose(color_transforms + scale_transforms) if image_set == 'val': return T.MotCompose([ T.MotRandomResize([800], max_size=1536), normalize, ]) raise ValueError(f'unknown {image_set}')
def make_transforms_for_crowdhuman(image_set, args=None): normalize = T.MotCompose([ T.MotToTensor(), T.MotNormalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]) ]) scales = [608, 640, 672, 704, 736, 768, 800, 832, 864, 896, 928, 960, 992] if image_set == 'train': return T.MotCompose([ T.MotRandomHorizontalFlip(), T.FixedMotRandomShift(bs=1), T.MotRandomSelect( T.MotRandomResize(scales, max_size=1536), T.MotCompose([ T.MotRandomResize([400, 500, 600]), T.FixedMotRandomCrop(384, 600), T.MotRandomResize(scales, max_size=1536), ]) ), normalize, ]) if image_set == 'val': return T.MotCompose([ T.MotRandomResize([800], max_size=1333), normalize, ]) raise ValueError(f'unknown {image_set}')
def make_detmot_transforms(image_set, args=None): normalize = T.MotCompose([ T.MotToTensor(), T.MotNormalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]) ]) scales = [480, 512, 544, 576, 608, 640, 672, 704, 736, 768, 800] if image_set == 'train': color_transforms = [] scale_transforms = [ T.MotRandomHorizontalFlip(), T.MotRandomResize(scales, max_size=1333), normalize, ] return T.MotCompose(color_transforms + scale_transforms) if image_set == 'val': return T.MotCompose([ T.MotRandomResize([800], max_size=1333), normalize, ]) raise ValueError(f'unknown {image_set}')