Replication Recommendations #19
Description
Hi all,
Thank you for your interest in our work. Due to time constraints during the development of this public code base, it is not fully stable or optimised, particularly for training. We recommend implementing your own version based on our model details and code provided. There is big room for further exploration. You may also develop your own version of the point-based evaluation code, from scratch or based on AB3DMOT. In addition, as this work is patented, we are limited in the amount of information we can share in response to your requests.
We respect all your evaluation methods and results and appreciate your understanding of these limitations. While evaluating, please keep in mind that bicycle and cyclist (person) objects are separately annotated in the ground truth in the VoD dataset. Additionally, remember to ignore objects with fewer than N points in both the ground truth and inference results.
Best regards,
Z