Tian Zhi

@hello-piger yes, you can use the command line, but you cannot benefit from this distributed training.

@Hwang64 It is here https://github.com/HRNet/HRNet-FCOS.

@Hwang64 It is not normal. We normally initialize the models with ImageNet pre-trained weights.

@marico2020 It seems you did not compile the project.

@hheavenknowss Please try to clip the gradients.

@maojiaoli It can be downloaded here https://dl.fbaipublicfiles.com/detectron/coco/coco_annotations_minival.tgz

Thank you for your questions. > for a specific prediction, the iou is calculated over all GT bboxes with the same class as this prediction and the maximum is chosen....

@Emma0118 We tried some initial experiments, and FCOS can achieve on-par results with RetinaNet on VOC.

We used the training settings in detectron2. https://github.com/facebookresearch/detectron2/blob/master/configs/PascalVOC-Detection/faster_rcnn_R_50_FPN.yaml But they seem to be similar.

@EchoTang123 Do you follow the installation instructions to compile the project?