LaCandela

Could you clarify what happens in build_targets() function?

2

Could you explain a bit how does the anchor box to ground truth box maching is performed in utils.loss.build_targets() ? [https://github.com/WongKinYiu/yolor/blob/be7da6eba2f612a15bf462951d3cdde66755a180/utils/loss.py#L152](url) It seems that somehow the aspect ratio is used...

Convolutional layers with groups cannot be converted properly

1

I tried to convert a keras model that contains 2D group convolutions and in the resulting ONNX file the group parameter is always 1 even though it should be different....

Question: reducing blinking and bounding box jitter

4

To what extent can we expect this method to reduce bounding box blinking and jittering? What parameters can I try to improve the bounding box stabilization effect? So far I...

Huge memory consumption during training

1

I noticed that during training the process consumes lots of CPU memory. For example, if I use 8 workers with batch size of 45 the memory consumption is more than...

Question concerning temporal fusion

5

I understand that Fast-BEV uses 4 consecutive frames to improve performance. Do you correct the BEV features of the past frames before temporal fusion (concatenation)? I would naively think that...

ONNX model export?

2

Has anyone tried to export any of the FB-OCC or FB-BEV model to ONNX format? Is it even possible?

ONNX export?

1

I am wondering if it is possible to export the model in ONNX format. Is it even possible? Is there any unsupported operation? Has anyone tried it before? Thanks!

Cannot reproduce published performance after training

4

I am trying to reproduce the published model quality but I couldn't so far. I retrained an m0 model version and got only 0.215 NDS and 0.04 mAP instead of...