Hi @Tianxiaomo
Thank you for the code. I would like to consult a question.
As far as I know, yolov4 still uses the anchor-based object-detection approach like yolov3. I have looked through the train.py script in this repository and it seems to me that you use fixed anchors in Yolo_loss function?
Is yolov4 neglecting the need of generating dataset-specific anchors? Or else, do you intend to add k-means-based anchor generation function?
TA-Robot
commented
3 years ago
Hi @Tianxiaomo Thank you for the code. I would like to consult a question. As far as I know, yolov4 still uses the anchor-based object-detection approach like yolov3. I have looked through the train.py script in this repository and it seems to me that you use fixed anchors in Yolo_loss function? Is yolov4 neglecting the need of generating dataset-specific anchors? Or else, do you intend to add k-means-based anchor generation function?