-
Notifications
You must be signed in to change notification settings - Fork 84
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Wrong detection boxes with FCOS detector #75
Comments
Hi, I have tested on 1660ti and 2070s, with both PyTorch 1.8 and 1.9, TensorRT8.0.3, the code works on my side. |
I have the same Problem: #74 |
Hi, Attached are two Dockerfiles: 1 where the bug is present and 1 where the bug is NOT present. The only difference is the CUDA, pytorch and TensorRT versions. They are defined at the TOP of the Dockerfile. mmdet2trt_not_working.zip For reference, the versions that WORKS are the following:
Best regards! |
@grimoire Hi, Can you please tell which cuda version you are using with pytorch 1.8.0/1.9.0 and TensorRT 8.0.3? Is it CUDA 11.1? |
I can reproduce the error now after update the TensorRT version. And I will try to fix it ASAP. |
Let me know if I can help |
Hi, |
It fixes the problem! Thank you very much!
CUDA: 11.3 |
Describe the bug
When using demo/inference.py to perform inference with FCOS detector, the resulting detection boxes all have (x1, y1) coordinates equal to (0, 0). This leads to the following detections:
After investigations, we found that the bboxes computed here: https://github.com/grimoire/mmdetection-to-tensorrt/blob/master/mmdet2trt/models/dense_heads/fcos_head.py#L62 are corrects (x1 and y1 coordinates are usually not 0) but after the end of the loop here: https://github.com/grimoire/mmdetection-to-tensorrt/blob/master/mmdet2trt/models/dense_heads/fcos_head.py#L67, the (x1, y1) coordinates are all zeros except for the last scale.
Can you please have a look?
To Reproduce
python3 demo/inference.py base_path/demo/demo.jpg base_path/configs/fcos/fcos_r50_caffe_fpn_gn-head_1x_coco.py base_path/checkpoints/fcos_r50_caffe_fpn_gn-head_1x_coco-821213aa.pth /dev/shm/fcos.engine
enviroment:
Additional context
Executed on RTX 3060 for Laptop
The text was updated successfully, but these errors were encountered: