MobileNetV1, MobileNetV2, VGG based SSD/SSD-lite implementation in Pytorch 1.0 / Pytorch 0.4. Out-of-box support for retraining on Open Images dataset. ONNX and Caffe2 support. Experiment Ideas like CoordConv.
Currently images with all 0 labels are not used in loss calculation. So change made in box_utils.hard_negative_mining to accommodate images with all 0 labels. The idea is that 0 is considered only in classification loss, not in bbox regression loss.
A batch should contain at least one positive label.
Currently images with all 0 labels are not used in loss calculation. So change made in box_utils.hard_negative_mining to accommodate images with all 0 labels. The idea is that 0 is considered only in classification loss, not in bbox regression loss. A batch should contain at least one positive label.