def job(fn, gpuid):
#flow = flowlib.read_flow(fn)
#newflow = F.interpolate(torch.tensor(flow).to('cuda:{0}'.format(gpuid)).permute(2,0,1).unsqueeze(0), scale_factor=224/min(flow.shape[:-1])).squeeze()
newfn = fn.replace('scenes', 'scenes_small')
#if os.path.exists(newfn): return 0
img = Image.open(fn).convert('RGB')
os.makedirs(os.path.dirname(newfn), exist_ok=True)
img = Resize(224)(img)
img.save(newfn)
return 0
def main():
CUDA = torch.cuda.is_available()
opts = opt()
pose_model = model.poseprosalnet(mpii_dataset.KEYPOINT_NAMES,
mpii_dataset.EDGES, opts.local_grid_size,
opts.insize)
pose_model.load_state_dict(torch.load(opts.checkpoint))
pil_image = Image.open(opts.img)
pil_image = Resize(opts.insize)(pil_image)
img = ToTensor()(pil_image)
img = Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])(img)
img = img.unsqueeze(0)
pose_model.eval()
print(mpii_dataset.KEYPOINT_NAMES)
with torch.no_grad():
if CUDA:
pose_model.cuda()
img = img.cuda()
pre = pose_model(img)
humans = utils.get_humans_by_feature(
pre,
opts.insize,
pose_model.outsize,
opts.local_grid_size,
)
# print(humans[0])
# print(humans[1])
pil_image = utils.draw_humans(mpii_dataset.KEYPOINT_NAMES,
mpii_dataset.EDGES,
pil_image,
humans,
visbbox=False)
pil_image.save("result.png", "PNG")
args.weights = './model_zoo/espnetv2/espnetv2_s_2.0_imagenet_224x224.pth'
from PIL import Image
model = EESPNet(args)
model.eval()
weigth_dict = torch.load(args.weights)
model.load_state_dict(weigth_dict)
rgb_img = Image.open('../train_140_im/MP_0001.tiff.jpg').convert('RGB')
from torchvision.transforms import CenterCrop, ToTensor, Normalize, Resize
from data_loader.melanoma_classification import MEAN, STD
crop_image = Resize(size=args.im_size)(rgb_img)
crop_image.save('../temp/rgb.jpg')
crop_image = ToTensor()(crop_image)
crop_image_norm = Normalize(mean=MEAN, std=STD)(crop_image)
crop_image_norm = crop_image_norm.unsqueeze(0)
_, outputs = model(crop_image_norm)
import matplotlib.pyplot as plt
for i in range(len(outputs)):
pred = outputs[i]
pred = F.softmax(pred, dim=1).squeeze(0)
pred = pred.max(0)[0]
h, w = pred.size()
pred = pred.cpu().detach().numpy()
rgb_paths = [
root + 'train_rgb/' + d for d in os.listdir(root + 'train_rgb/')
]
augmentation = Compose([RandomCrop((420, 560)), RandomCropRotate(10)])
counter = 3
while counter < 6:
counter += 1
c = 0
for path in rgb_paths:
c += 1
print('%d %d/%d' % (counter, c, len(rgb_paths)))
img = Image.open(path)
depth = Image.open(path.replace('rgb', 'depth'))
rgb, depth = np.array(img), np.array(depth)[:, :, None]
rgb, depth = augmentation((rgb, depth))
depth = depth.squeeze(-1)
rgb, depth = Image.fromarray(rgb), Image.fromarray(depth)
rgb, depth = Resize((420, 560))(rgb), Resize((420, 560))(depth)
rgb = ColorJitter(brightness=0.2,
contrast=0.2,
saturation=0.2,
hue=0.1)(rgb)
rgb.save(root + 'aug_rgb/' + path.split('/')[-1].split('.')[0] +
str(counter) + '.png')
depth.save(
root + 'aug_depth/' +
path.replace('rgb', 'depth').split('/')[-1].split('.')[0] +
str(counter) + '.png')
