Closed heyraselyn closed 5 years ago
The difference is clear from 0.5 and 0.75. The paper's results were generated with caffe implementation of SSD, which used some different parameters. I think it is due to different beta value in l1 loss function of pytorch and caffe.
Hi. I'm just confused about your reported results. It says in your paper that using RGB images only, the results are as follow: @0.2: 69.8, @0.5:40.9, @.75: 15.5, @Avg: 18.7 But from the table in this repo: This implentation[4] RGB: @0.2: 72.08, @0.5: 40.59, @0.75: 14.06, @Avg: 18.48
Why is that so? Thank you.