Pretrained model for surface normal estimation

[Ainaz99]

Hi,

Is the pretrained model best_model_iter_249000.bin released for surface normal estimation trained on SNOW or OASIS dataset? According to the args.pkl file provided, the model is trained on SNOW:

{'train_file': '/n/fs/vl/datasets/SNOW/SNOW_Toolkit/SNOW_train_release.csv', 
'valid_file': '/n/fs/vl/datasets/SNOW/SNOW_Toolkit/SNOW_val_release.csv', 
'dataset_name': 'SNOWDataset', 'model_name': 'NIPSSurface', 'loss_name': 'CosineAngularLoss',
'in_coord_sys': 'SNOW', 'out_coord_sys': 'SNOW', 'num_iters': 500000, 'num_epoches': 100000, 'batch_size': 12, 
'model_save_interval': 5000, 'model_eval_interval': 3000, 'learning_rate': 0.0001, 'n_GPUs': 1,
'num_loader_workers': 2, 'pretrained_model': None, 'pretrained_optimizer': None, 'data_aug': False}

[wfchen-umich]

Hi, This model is indeed trained on the OASIS dataset, not the SNOW dataset. Sorry about the confusion. Please feel free to ignore those info stored in args.pkl.

[Ainaz99]

Thank you.