search

ykotseruba / PedestrianActionBenchmark

Code and models for the WACV 2021 paper "Benchmark for evaluating pedestrian action prediction"
https://openaccess.thecvf.com/content/WACV2021/papers/Kotseruba_Benchmark_for_Evaluating_Pedestrian_Action_Prediction_WACV_2021_paper.pdf
MIT License
51 stars 17 forks source link

pose dataset #28

Open lcs1027 opened 5 days ago

lcs1027 commented 5 days ago

Hi Yulia,

I notice that in pose_set01.pkl of JAAD, there are pose data for 294 videos. Are these train (177) + test (117)? I am wondering have you provided pose data for val subset? In the code, data_val can read pose data, but I am not sure why in this case there are only data for 294 videos in pose_set01.pkl.

Thanks!

ykotseruba commented 2 days ago

Hi, pose data is not available for some videos because the pedestrians are either too far or occluded. In that case, a dummy value is substituted. Missing pose data is not related to the train/val/test split. The train/val/test sets in JAAD are formed by randomly selecting 188/32/126 video IDs (https://github.com/ykotseruba/JAAD/tree/JAAD_2.0/split_ids/all_videos), so every set has some missing pose data. Hope this answers your concern.