New features:
- A powerful convnext-tiny driven YOLOv7 model, mAP 43 without any augmentation training;
- Convnext-tiny YOLOv7 (AKA yoloformer) get mAP-s 26 which 10 AP higher than **MaskRCNN**!;
- YOLO2go proposed, will make a small weight size YOLO model (in mobilenetv3 or efficicientnetv2 backbone), will release in next version.
v.0.0.3
Since YOLOv7 not just yolo, it aims at a mature modeling tool based on detectron2, so I will continue add more models into it. It now also received a lot of PR from community. Let's build YOLOv7 togather and make a lot of experiment on it.
Current is a Beta release, it becoming close to a Stable release version, we are keep on doing on that.
- YOLOX-Pose model added, now you should able to training YOLOx with Pose head;
- SOLOv2 training support, now you can using YOLOv7 training SOLOv2;
- Sparseinst training support, now you can train sparseinst, we also support ONNX exportation;
- Some bugs fixes in sparseinst;
Feature you will see in the next release:
- Int8 quantization on Sparseinst, **CPU Realtime** instance segmentation, just from yolov7!
- TensorRT on SOLOv2 and Sparseinst;
- Pretrain model of YOLOX-Pose.
Just start and fork it! Join the community!