aim-uofa / AdelaiDet

AdelaiDet is an open source toolbox for multiple instance-level detection and recognition tasks.
https://git.io/AdelaiDet
Other
3.36k stars 643 forks source link

About some training and testing results of ABCNetV2 #474

Open wenston2006 opened 2 years ago

wenston2006 commented 2 years ago

I tried to train ABCNetV2 to detect and recognize the chars on Chinese stamps. I modify the _configs/ReCTS/v2_chn_attn_R50.yaml file, where I set:

_MIN_SIZE_TRAIN: (128, 256, 384, 512, 640, 768) MAX_SIZE_TRAIN: 1024 MIN_SIZE_TEST: 128 MAX_SIZETEST: 1024

After training 19999 iterations, I test the models over some stamps, the result is as follows:

1

My question is: Is the result above correct?

wenston2006 commented 2 years ago

I mean: is there any problem with my training process?

Yuliang-Liu commented 2 years ago

@wenston2006 MIN_SIZE_TRAIN and MIN_SIZE_TEST should not be too small. You may follow the original setting

tkone2018 commented 2 years ago

你好,我在测试推理的时候遇到一个错,可以帮忙看下吗 Traceback (most recent call last): File "demo/demo.py", line 75, in demo = VisualizationDemo(cfg) File "/root/code/adet/demo/predictor.py", line 40, in init self.predictor = DefaultPredictor(cfg) File "/root/code/detectron2/detectron2/engine/defaults.py", line 182, in init self.model = build_model(self.cfg) File "/root/code/detectron2/detectron2/modeling/meta_arch/build.py", line 21, in build_model model = META_ARCH_REGISTRY.get(meta_arch)(cfg) File "/root/code/adet/adet/modeling/one_stage_detector.py", line 72, in init super().init(cfg) File "/root/code/detectron2/detectron2/config/config.py", line 147, in wrapped explicit_args = _get_args_from_config(from_config_func, *args, *kwargs) File "/root/code/detectron2/detectron2/config/config.py", line 180, in _get_args_from_config ret = from_config_func(args, **kwargs) File "/root/code/detectron2/detectron2/modeling/meta_arch/rcnn.py", line 75, in from_config backbone = build_backbone(cfg) File "/root/code/detectron2/detectron2/modeling/backbone/build.py", line 31, in build_backbone backbone = BACKBONE_REGISTRY.get(backbone_name)(cfg, input_shape) File "/root/code/adet/adet/modeling/backbone/bifpn.py", line 389, in build_fcos_resnet_bifpn_backbone backbone = BiFPN( File "/root/code/adet/adet/modeling/backbone/bifpn.py", line 331, in init self.repeated_bifpn.append(SingleBiFPN( File "/root/code/adet/adet/modeling/backbone/bifpn.py", line 146, in init raise NotImplementedError NotImplementedError 谢谢