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
56 stars 17 forks source link

The updated evaluation results #13

Closed d-zh closed 2 years ago

d-zh commented 3 years ago

Hi, As you mentioned in https://github.com/ykotseruba/PedestrianActionBenchmark/issues/3#issuecomment-782693588, the number of samples in JAAD decreased. Could you please provide the updated evaluation results in paper Table 2, 3? Thanks

ykotseruba commented 3 years ago

Hi @d-zh, Table 2 is unchanged, the variation was very insignificant and within the range that we normally got when retraining the models multiple times. In Table 3 only the PCPA with both attention modules was hit by the changes because it is the largest model. On JAAD all the best model we have is acc=0.83 auc=0.83 f1=0.64, on JAADbeh acc=0.56 auc=0.56 and f1=0.68. On PIE the results are unchanged.

ykotseruba commented 2 years ago

Closed due to inactivity