search

hasanirtiza / Pedestron

[Pedestron] Generalizable Pedestrian Detection: The Elephant In The Room. @ CVPR2021
https://openaccess.thecvf.com/content/CVPR2021/papers/Hasan_Generalizable_Pedestrian_Detection_The_Elephant_in_the_Room_CVPR_2021_paper.pdf
Apache License 2.0
682 stars 159 forks source link

result has no bbox #163

Closed decajcd closed 1 year ago

decajcd commented 1 year ago

I run 'python tools/train.py configs/elephant/cityperson/cascade_hrnet.py' to train,and in configs/elephant/cityperson/cascade_hrnet.py,img_scale=[(1216, 608), (2048, 1024)] is modified to [(608, 304), (1024, 512)],load_from='epoch_5.pth'.After one round of training,when i run python demo.py,the result has no bbox,just oral image

hasanirtiza commented 1 year ago

With this information cant really comment on anything. How was your training log. Did you first simply try running the epoch5.pth on your images ?

decajcd commented 1 year ago

With this information cant really comment on anything. How was your training log. Did you first simply try running the epoch5.pth on your images ? epoch5.pth has bbox

decajcd commented 1 year ago

With this information cant really comment on anything. How was your training log. Did you first simply try running the epoch5.pth on your images ? when I trainthe model,the acc is about 98%,but when I run demo,it has no box

hasanirtiza commented 1 year ago

Its somewhat of a strange behavior, may be let it run for few epochs. Secondly, you can also try to reduce the learning rate. Lastly, as a sanity check, make sure your annotations are correct.

decajcd commented 1 year ago

Its somewhat of a strange behavior, may be let it run for few epochs. Secondly, you can also try to reduce the learning rate. Lastly, as a sanity check, make sure your annotations are correct.

After modify img_scale,how to change bbox location?