YoloV8训练单通道数据方法+单通道Onnx模型转换

1. ultralytics/cfg/models/v8/yolov8-xxx.yaml

添加ch: 1

2. ultralytics/utils/checks.py

注释assert amp_allclose(YOLO("yolov8n.pt"), im)

3. ultralytics/nn/tasks.py

添加：batch['img'] = batch['img'][:,:1,:,:]

4. ultralytics/nn/autobackend.py

修改：def warmup(self, imgsz=(1, 1, 640, 640)):

5. ultralytics/engine/validator.py

修改：model.warmup(imgsz=(1 if pt else self.args.batch, 1, imgsz, imgsz))

添加：batch["img"] = batch["img"][:,:1,:,:]

6. ultralytics/engine/predictor.py

修改：self.model.warmup(imgsz=(1 if self.model.pt or self.model.triton else self.dataset.bs, 1, *self.imgsz))

训练结束后如何转出onnx模型

ultralytics/engine/exporter.py

修改im = torch.zeros(self.args.batch, 3, *self.imgsz).to(self.device)改为im = torch.zeros(self.args.batch, 1, *self.imgsz).to(self.device)

训练结束后如何测试pt模型

找到对应任务的ultralytics/models/yolo/classify(自己任务)/predict.py
添加：img = img[:,:1,:,:]

参考

github issue

补充，发现推理时会提示输入尺寸是三通道，在ultralytics/models/yolo/classify/predict.py中加入img = img[:,:1,:,:]即可以推理成功

或者更好的方法应该是ultralytics/engine/predictor.py添加
img = img[:,:1,:,:]