【RF-DETR】目标检测新王者 从训练到部署

https://rfdetr.roboflow.com/learn/train/#fine-tuning
https://github.com/roboflow/rf-detr

https://github.com/roboflow/rf-detr

pip install rfdetr -i https://mirrors.aliyun.com/pypi/simple

from rfdetr import RFDETRNano
import pycocotools.coco as coco
import copy
original_loadRes = coco.COCO.loadRes
def patched_loadRes(self, resFile):
    try:
        return original_loadRes(self, resFile)
    except Exception as e:
        # Create a minimal COCO result object
        res = coco.COCO()
        res.dataset = copy.deepcopy(self.dataset)
        # Add minimal required structure
        if 'annotations' not in res.dataset:
            res.dataset['annotations'] = []
        return res
# Apply patch
coco.COCO.loadRes = patched_loadRes
model = RFDETRNano()
model.train(
    dataset_dir="barray_det_ds",
    epochs=100,
    batch_size=4,
    grad_accum_steps=4,
    lr=1e-4,
    output_dir="output"
)

from rfdetr import RFDETRNano
model = RFDETRNano(pretrain_weights="rf-detr-nano.pth")
model.export()