srama2512/sidekicks

Sidekick Policy Learning for Active Visual Exploration (ECCV 2018)

sidekicks's Issues

Training with multi-step

Hi Santhosh,

I could train the with a single step, but I cannot train the model with multiple steps (--T 4):
python main.py --T 4 --training_setting 1 --epochs 1000 \ --save_path saved_models/sun360/ltla/ \ --load_model models/sun360/one-view.net

which stops in:

"misc/State.py", line 180, in rotate
delta = [list(self.act_to_delta[act[i]]) for i in range(act.shape[0])]

Would you please give me any suggestions regarding this?

Clean up code-base and naming conventions

Some scripts do not have documentation yet. Some have misleading naming conventions, for example, create_utility_based_rewards.py is used for ours-demo-scores and create_one_view_scores.py is used for ours-rew-scores.