Ming Kang

"In SSFF, The P3, P4, and P5 feature maps are normalized to the same size, upsampled, and then stacked together as input to a 3D convolution to combine multiscale features."...

Thanks for pointing this problem out. To clarify the expression, we replace `scale-invariant features' with `aspect ratio invariant features'. As you know, each learned filter of Convolutional neural networks (CNNs),...

Of course. ASF-YOLO can apply to both object detection and instance segmentation.

The mAP is the most commonly used metric for object detection and instance segmentation, which builds on top of IoU and the precision-recall curve. mIoU is an evaluation metric for...

I suppose you'd better post your issue on the [PointRend](https://github.com/facebookresearch/detectron2/tree/master/projects/PointRend) or [YOLOv8](https://github.com/ultralytics/ultralytics) repository.

Please download the 3-head version RCS-YOLO from: https://github.com/mkang315/RCS-YOLO/blob/v1.1/cfg/training/rcs3-yolo.yaml Thank you for your support!

Modified. Thank you!

Sorry for that. You may download ASF-YOLO v1.0 if you don't need the file yolov5l-seg.pt. Thank you for being so considerate.

That error demonstrated your network structure was not reasonable. Please see [repvgg-csp.yaml](https://github.com/mkang315/RCS-YOLO/blob/main/cfg/ablation/repvgg-csp.yaml). FYI.

I'm sorry that the names of some modules in the code were changed in the manuscript. The attention_model is in [./models/common.py](https://github.com/mkang315/ASF-YOLO/blob/main/models/common.py). Please refer to class attention_model(nn.Module). Thank you.