PyTorchClassification/data_loader.py [28:165]: - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - def show_image(input, bbox=None): import matplotlib.pyplot as plt img = np.transpose(input.data.numpy(),[1, 2, 0]) plt.imshow(img) if (bbox): x = [bbox[0], bbox[0], bbox[0]+bbox[2], bbox[0]+bbox[2], bbox[0]] y = [bbox[1], bbox[1]+bbox[3], bbox[1]+bbox[3], bbox[1], bbox[1]] plt.plot(x, y) plt.show() def deploy_model(filein, fileout): """ Loads a model from a checkpoint, then re-saves it in a format that is more practical for deployment for inference-only applications. """ print("=> deploying checkpoint '{}'".format(filein)) checkpoint = torch.load(filein, map_location=lambda storage, loc: storage) deploy_checkpoint = { 'epoch' : checkpoint['epoch'], 'state_dict': checkpoint['state_dict'], 'classnames' : checkpoint['classnames'], 'model_type' : checkpoint['model_type']} torch.save(deploy_checkpoint, fileout) def save_model(state, is_best, filename='checkpoint.pth.tar'): """ Saves a model to a checkpoint. """ torch.save(state, filename) if is_best: print("\tSaving new best model") shutil.copyfile(filename, 'model_best.pth.tar') def load_model(filename, useGPU=True): """ Loads a model from a checkpoint. """ if os.path.isfile(filename): print("=> loading checkpoint '{}'".format(filename)) if useGPU: cuda_device = torch.cuda.current_device() checkpoint = torch.load(filename, map_location=lambda storage, loc: storage.cuda(cuda_device)) else: checkpoint = torch.load(filename, map_location=lambda storage, loc: storage) start_epoch = checkpoint['epoch'] if 'epoch' in checkpoint else 0 best_prec1 = checkpoint['best_prec1'] if 'best_prec1' in checkpoint else 0 best_prec3 = checkpoint['best_prec3'] if 'best_prec3' in checkpoint else 0 best_prec5 = checkpoint['best_prec5'] if 'best_prec5' in checkpoint else 0 state_dict = checkpoint['state_dict'] classnames = checkpoint['classnames'] model_type = checkpoint['model_type'] print('Loaded %d classes' % len(classnames)) new_state_dict = OrderedDict() for k, v in state_dict.items(): module = k[0:7] # check for 'module.' of dataparallel name = k[7:] # remove 'module.' of dataparallel if k[:7] == 'module.': k = k[7:] if k[:2] == '1.': k = k[2:] if k[:6] == 'model.': k = k[6:] new_state_dict[k] = v #print("%s" % (k)) model_dict = new_state_dict optimizer_dict = checkpoint['optimizer'] if 'optimizer' in checkpoint else None print("=> loaded checkpoint '{}' (epoch {})" .format(filename, start_epoch)) data.best_prec1 = best_prec1 data.best_prec3 = best_prec3 data.best_prec5 = best_prec5 data.start_epoch = start_epoch data.classnames = classnames data.model_dict = model_dict data.optimizer_dict = optimizer_dict data.model_type = model_type return data else: print("=> no checkpoint found at '{}'".format(filename)) # ...def load_model(filename, useGPU=True) class ImageLoader(): def __init__(self, image_sizes): # The largest image size is used as target size in preprocessing # The scaling to the proper size should be done within the model self.im_size = [max(image_sizes), max(image_sizes)] self.mu_data = [0.5, 0.5, 0.5] self.std_data = [0.5, 0.5, 0.5] self.brightness = 0.4 self.contrast = 0.4 self.saturation = 0.4 self.hue = 0.25 # augmentations self.center_crop = transforms.CenterCrop((self.im_size[0], self.im_size[1])) self.random_crop = transforms.RandomCrop((self.im_size[0], self.im_size[1]),pad_if_needed=True) self.scale_aug = transforms.RandomResizedCrop(size=self.im_size[0]) self.flip_aug = transforms.RandomHorizontalFlip() self.flip = transforms.RandomHorizontalFlip(1.0) self.color_aug = transforms.ColorJitter(self.brightness, self.contrast, self.saturation, self.hue) self.tensor_aug = transforms.ToTensor() self.norm_aug = transforms.Normalize(mean=self.mu_data, std=self.std_data) self.resize = transforms.Resize((self.im_size[0], self.im_size[1])) self.resize_for_crop = transforms.Resize((int(1.14 * self.im_size[0]), int(1.14 * self.im_size[1]))) self.multi_crop = transforms.TenCrop((self.im_size[0], self.im_size[1])) def load_image(self, path): - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - PyTorchClassification/data_loader_cv.py [29:166]: - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - def show_image(input, bbox=None): import matplotlib.pyplot as plt img = np.transpose(input.data.numpy(),[1, 2, 0]) plt.imshow(img) if (bbox): x = [bbox[0], bbox[0], bbox[0]+bbox[2], bbox[0]+bbox[2], bbox[0]] y = [bbox[1], bbox[1]+bbox[3], bbox[1]+bbox[3], bbox[1], bbox[1]] plt.plot(x, y) plt.show() def deploy_model(filein, fileout): """ Loads a model from a checkpoint, then re-saves it in a format that is more practical for deployment for inference-only applications. """ print("=> deploying checkpoint '{}'".format(filein)) checkpoint = torch.load(filein, map_location=lambda storage, loc: storage) deploy_checkpoint = { 'epoch' : checkpoint['epoch'], 'state_dict': checkpoint['state_dict'], 'classnames' : checkpoint['classnames'], 'model_type' : checkpoint['model_type']} torch.save(deploy_checkpoint, fileout) def save_model(state, is_best, filename='checkpoint.pth.tar'): """ Saves a model to a checkpoint. """ torch.save(state, filename) if is_best: print("\tSaving new best model") shutil.copyfile(filename, 'model_best.pth.tar') def load_model(filename, useGPU=True): """ Loads a model from a checkpoint. """ if os.path.isfile(filename): print("=> loading checkpoint '{}'".format(filename)) if useGPU: cuda_device = torch.cuda.current_device() checkpoint = torch.load(filename, map_location=lambda storage, loc: storage.cuda(cuda_device)) else: checkpoint = torch.load(filename, map_location=lambda storage, loc: storage) start_epoch = checkpoint['epoch'] if 'epoch' in checkpoint else 0 best_prec1 = checkpoint['best_prec1'] if 'best_prec1' in checkpoint else 0 best_prec3 = checkpoint['best_prec3'] if 'best_prec3' in checkpoint else 0 best_prec5 = checkpoint['best_prec5'] if 'best_prec5' in checkpoint else 0 state_dict = checkpoint['state_dict'] classnames = checkpoint['classnames'] model_type = checkpoint['model_type'] print('Loaded %d classes' % len(classnames)) new_state_dict = OrderedDict() for k, v in state_dict.items(): module = k[0:7] # check for 'module.' of dataparallel name = k[7:] # remove 'module.' of dataparallel if k[:7] == 'module.': k = k[7:] if k[:2] == '1.': k = k[2:] if k[:6] == 'model.': k = k[6:] new_state_dict[k] = v #print("%s" % (k)) model_dict = new_state_dict optimizer_dict = checkpoint['optimizer'] if 'optimizer' in checkpoint else None print("=> loaded checkpoint '{}' (epoch {})" .format(filename, start_epoch)) data.best_prec1 = best_prec1 data.best_prec3 = best_prec3 data.best_prec5 = best_prec5 data.start_epoch = start_epoch data.classnames = classnames data.model_dict = model_dict data.optimizer_dict = optimizer_dict data.model_type = model_type return data else: print("=> no checkpoint found at '{}'".format(filename)) # ...def load_model(filename, useGPU=True) class ImageLoader(): def __init__(self, image_sizes): # The largest image size is used as target size in preprocessing # The scaling to the proper size should be done within the model self.im_size = [max(image_sizes), max(image_sizes)] self.mu_data = [0.5, 0.5, 0.5] self.std_data = [0.5, 0.5, 0.5] self.brightness = 0.4 self.contrast = 0.4 self.saturation = 0.4 self.hue = 0.25 # augmentations self.center_crop = transforms.CenterCrop((self.im_size[0], self.im_size[1])) self.random_crop = transforms.RandomCrop((self.im_size[0], self.im_size[1]),pad_if_needed=True) self.scale_aug = transforms.RandomResizedCrop(size=self.im_size[0]) self.flip_aug = transforms.RandomHorizontalFlip() self.flip = transforms.RandomHorizontalFlip(1.0) self.color_aug = transforms.ColorJitter(self.brightness, self.contrast, self.saturation, self.hue) self.tensor_aug = transforms.ToTensor() self.norm_aug = transforms.Normalize(mean=self.mu_data, std=self.std_data) self.resize = transforms.Resize((self.im_size[0], self.im_size[1])) self.resize_for_crop = transforms.Resize((int(1.14 * self.im_size[0]), int(1.14 * self.im_size[1]))) self.multi_crop = transforms.TenCrop((self.im_size[0], self.im_size[1])) def load_image(self, path): - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -