Closed dpsdev14091998 closed 4 years ago
Hey @ifzhang Which one is usually better on a human-based dataset?
@apekshapriya I test them on the test set of MOT17 dataset and find DLA-34 has better results. The deformable convolution in DLA-34 plays an important role. HRNet-W18 has better performance on small objects.
@ifzhang Thanks!
Yes, it will change. The MOTA is affected by the training model and testing parameters.