search

facebookresearch / AVT

Code release for ICCV 2021 paper "Anticipative Video Transformer"
Apache License 2.0
152 stars 28 forks source link

Long-term anticipation #40

Open RodinIvan opened 2 years ago

RodinIvan commented 2 years ago

Hi, thanks for repo! Excuse me, could you please tell how to perform long-term anticipation as showed in the paper? Have I missed this part in code, or it's not implemented here yet?

rohitgirdhar commented 2 years ago

Hi @RodinIvan thanks for your interest in this work! I don't have a config for it since we only used it for some visualizations, however you should be able to get that output by setting the following and running the evaluation with a breakpoint around here

+model.future_predictor.avg_last_n=-1
data_eval.num_frames=3  # To give it only 3 frames to start predicting from
+model.future_predictor.output_len=10  # To predict 10 steps into the future instead of 1
+dataset_eval.load_long_term_future_labels=${model.future_predictor.output_len}

It will need a bit more code to generate the figures shown in the paper but I haven't gotten around to add that yet