model_ft.fc = nn.Linear(num_ftrs, num_classes)
input_size = 299
else:
print("Invalid model name, exiting...")
exit()
return model_ft, input_size
pickle_files = [
os.path.join(data_dir, f) for f in os.listdir(data_dir)
if os.path.isfile(os.path.join(data_dir, f))
]
print(pickle_files)
data_loader_train, data_loader_val = get_dataloader(batch_size=batch_size,
VG_DIR=vg_dir)
# Create training and validation dataloaders
dataloaders_dict = {'train': data_loader_train, 'val': data_loader_val}
num_classes = data_loader_train.dataset.num_objects
# Initialize the model for this run
model_ft, input_size = initialize_model(model_name,
num_classes,
feature_extract,
use_pretrained=True)
# Print the model we just instantiated
# print(model_ft)
# Detect if we have a GPU available
_, z_rand_rec, _ = self.crop_encoder(crops_rand, objs)
crops_input_rec = crop_bbox_batch(img_rec, boxes, obj_to_img,
self.obj_size)
return crops_input, crops_input_rec, crops_rand, img_rec, img_rand, mu, logvar, z_rand_rec
if __name__ == '__main__':
from data.vg_custom_mask import get_dataloader
device = torch.device('cuda:0')
z_dim = 8
batch_size = 4
train_loader, _ = get_dataloader(batch_size=batch_size)
vocab_num = train_loader.dataset.num_objects
# test CropEncoder
# model = CropEncoder(class_num=vocab_num).to(device)
#
# for batch in train_loader:
# imgs, objs, boxes, masks, obj_to_img = batch
# imgs, objs, boxes, masks, obj_to_img = imgs.to(device), objs.to(device), boxes.to(device), masks.to(device), obj_to_img.to(device)
#
# crops = crop_bbox_batch(imgs, boxes, obj_to_img, 32)
# outputs = model(crops, objs)
#
# for output in outputs:
# print(output.shape)
#
# Handle the auxilary net
num_ftrs = model_ft.AuxLogits.fc.in_features
model_ft.AuxLogits.fc = nn.Linear(num_ftrs, num_classes)
# Handle the primary net
num_ftrs = model_ft.fc.in_features
model_ft.fc = nn.Linear(num_ftrs, num_classes)
input_size = 299
else:
print("Invalid model name, exiting...")
exit()
return model_ft, input_size
data_loader_train, data_loader_val = get_dataloader(batch_size=batch_size, attribute_embedding=106)
iters_per_epoch = {"train": len(data_loader_train)//batch_size, "val": len(data_loader_val)//batch_size}
# Create training and validation dataloaders
dataloaders_dict = {'train': data_loader_train, 'val': data_loader_val}
num_classes = data_loader_train.dataset.num_objects
# Initialize the model for this run
model_ft, input_size = initialize_model(model_name, num_classes, feature_extract, use_pretrained=True)
# Print the model we just instantiated
print(model_ft)
# Detect if we have a GPU available