NagabhushanSN95
commented
5 years ago I tried skipping this step. I got the below error
CUDA_VISIBLE_DEVICES=0 python scripts/generate.py --input_dir data/bair --dataset_hparams sequence_length=30 --checkpoint pretrained_models/bair_action_free/ours_savp --mode test --results_dir results_test_samples/bair_action_free loading options from checkpoint pretrained_models/bair_action_free/ours_savp ----------------------------------- Options ------------------------------------ batch_size = 8 checkpoint = pretrained_models/bair_action_free/ours_savp dataset = softmotion dataset_hparams = sequence_length=30 fps = 4 gif_length = None gpu_mem_frac = 0 input_dir = data/bair mode = test model = savp model_hparams = None num_epochs = 1 num_samples = None num_stochastic_samples = 5 output_gif_dir = results_test_samples/bair_action_free/ours_savp output_png_dir = results_test_samples/bair_action_free/ours_savp results_dir = results_test_samples/bair_action_free results_gif_dir = results_test_samples/bair_action_free results_png_dir = results_test_samples/bair_action_free seed = 7 ------------------------------------- End -------------------------------------- Traceback (most recent call last): File "scripts/generate.py", line 193, in <module> main() File "scripts/generate.py", line 105, in main hparams=args.dataset_hparams) File ".../Stochastic_Adversarial_Video_Prediction/video_prediction/video_prediction/datasets/softmotion_dataset.py", line 16, in __init__ super(SoftmotionVideoDataset, self).__init__(*args, **kwargs) File ".../Stochastic_Adversarial_Video_Prediction/video_prediction/video_prediction/datasets/base_dataset.py", line 242, in __init__ super(VideoDataset, self).__init__(*args, **kwargs) File ".../Stochastic_Adversarial_Video_Prediction/video_prediction/video_prediction/datasets/base_dataset.py", line 40, in __init__ raise FileNotFoundError("input_dir %s does not exist" % self.input_dir) FileNotFoundError: input_dir data/bair does not exist
In your doc, the first step when using a pretrained model is to download the dataset and preprocess it. Is this step really necessary? Is it possible to circumvent it? Because dataset is huge!