MedChaabane/DEFT

Occlusion Handling and Testing Phase.

Opened this issue · 1 comments

Thanks for publishing the source code. I would like to know where in the project I can find the codes for the following tasks:

  1. Handling Consistent Object Id before and After Occlusion.
  2. Embedding Extractor in Detector and Matching Head
  3. Memory Feature Embeddings, Motion Forecasting Module, and Detections-Tracks Association.

Hi,

  1. The code for handling occlusion can be found in tracker.py (src/lib/utils/tracker.py), you can read the online data association section in the paper for better understanding the idea (https://arxiv.org/abs/2102.02267)
  2. The code for the embedding extractor is AFE.py (src/lib/model/networks/AFE.py)
  3. Memory Feature Embedding can be found in FeatureRecorder class in tracker.py ; Motion forecasting module is src/lib/utils/tracking_utils/kalman_filter.py and src/lib/utils/tracking_utils/kalman_filter_lstm.py ; Detections-Tracks Association in the update function of the tracker class in src/lib/utils/tracker.py
    Thanks,