Open derrkater opened 2 years ago
The log shows that you used save_proposals in your config file, you could just modify the dataset config part based on https://github.com/PaddlePaddle/PaddleDetection/blob/release/2.5/configs/picodet/picodet_l_320_coco_lcnet.yml
@jerrywgz I've just modified config and used SniperCOCODataSet class to load the data but I'd think that proper behavior would be for COCODataSet to also work. Is it intended that it is missing the aforementioned method - init_anno_cropper
?
问题确认 Search before asking
Bug组件 Bug Component
Training
Bug描述 Describe the Bug
I'm trying default training command for picodet model with a custom coco-format data. It doesn't let me run my data in
COCODataSet
object:Traceback (most recent call last): File "tools/train.py", line 172, in <module> main() File "tools/train.py", line 168, in main run(FLAGS, cfg) File "tools/train.py", line 123, in run trainer = Trainer(cfg, mode='train') File "/home/PaddleDetection/ppdet/engine/trainer.py", line 191, in __init__ self._init_callbacks() File "/home/PaddleDetection/ppdet/engine/trainer.py", line 203, in _init_callbacks self._callbacks.append(SniperProposalsGenerator(self)) File "/home/PaddleDetection/ppdet/engine/callbacks.py", line 441, in __init__ self.dataset = self._create_new_dataset(ori_dataset) File "/home/PaddleDetection/ppdet/engine/callbacks.py", line 449, in _create_new_dataset dataset.init_anno_cropper() AttributeError: 'COCODataSet' object has no attribute 'init_anno_cropper'
复现环境 Environment
Docker image: registry.baidubce.com/paddlepaddle/paddle:2.3.2-gpu-cuda11.2-cudnn8 local cuda 11.4
Bug描述确认 Bug description confirmation
是否愿意提交PR? Are you willing to submit a PR?