RT-DETR主干网络预训练模型

[JianyuChenyyds]

如果想给RT-DETR换detection库的ppdet里的主干的话，主干网络的预训练权重在哪下载（backbone里没给出链接），是否可以用clas库里的预训练权重

[JianyuChenyyds]

Traceback (most recent call last): File "tools/train.py", line 209, in main() File "tools/train.py", line 205, in main run(FLAGS, cfg) File "tools/train.py", line 145, in run trainer = Trainer(cfg, mode='train') File "F:\RT-DETR\PaddleDetection-develop\ppdet\engine\trainer.py", line 115, in init self.model = create(cfg.architecture) File "F:\RT-DETR\PaddleDetection-develop\ppdet\core\workspace.py", line 255, in create cls_kwargs.update(cls.from_config(config, kwargs)) File "F:\RT-DETR\PaddleDetection-develop\ppdet\modeling\architectures\detr.py", line 58, in from_config neck = create(cfg['neck'], kwargs) if cfg['neck'] else None File "F:\RT-DETR\PaddleDetection-develop\ppdet\core\workspace.py", line 292, in create return cls(**cls_kwargs) File "F:\RT-DETR\PaddleDetection-develop\ppdet\modeling\transformers\hybrid_encoder.py", line 205, in init self._reset_parameters() File "F:\RT-DETR\PaddleDetection-develop\ppdet\modeling\transformers\hybrid_encoder.py", line 212, in _reset_parameters self.eval_size[1] // stride, self.eval_size[0] // stride, TypeError: unsupported operand type(s) for //: 'int' and 'NoneType'

[JianyuChenyyds]

Traceback (most recent call last): File "tools/train.py", line 209, in main() File "tools/train.py", line 205, in main run(FLAGS, cfg) File "tools/train.py", line 145, in run trainer = Trainer(cfg, mode='train') File "F:\RT-DETR\PaddleDetection-develop\ppdet\engine\trainer.py", line 115, in init self.model = create(cfg.architecture) File "F:\RT-DETR\PaddleDetection-develop\ppdet\core\workspace.py", line 255, in create cls_kwargs.update(cls.from_config(config, kwargs)) File "F:\RT-DETR\PaddleDetection-develop\ppdet\modeling\architectures\detr.py", line 58, in from_config neck = create(cfg['neck'], kwargs) if cfg['neck'] else None File "F:\RT-DETR\PaddleDetection-develop\ppdet\core\workspace.py", line 292, in create return cls(cls_kwargs) File "F:\RT-DETR\PaddleDetection-develop\ppdet\modeling\transformers\hybrid_encoder.py", line 205, in init** self._reset_parameters() File "F:\RT-DETR\PaddleDetection-develop\ppdet\modeling\transformers\hybrid_encoder.py", line 212, in _reset_parameters self.eval_size[1] // stride, self.eval_size[0] // stride, TypeError: unsupported operand type(s) for //: 'int' and 'NoneType'

出现这种问题怎么办

[lyuwenyu]

stride是None 看下你用的那个backbone有没有out_shape这个属性吧

[JianyuChenyyds]

stride是None 看下你用的那个backbone有没有out_shape这个属性吧

我是直接用的ppdet里面的主干，有out_shape

[JianyuChenyyds]

stride是None 看下你用的那个backbone有没有out_shape这个属性

好像是out_shape里面没有stride

[lyuwenyu]

可以自己手动加一下

[JianyuChenyyds]

可以自己手动加一下

一个是这种： @property def out_shape(self): return [ShapeSpec(channels=c) for c in self._out_channels] 一个是这种： @property def out_shape(self): return [ ShapeSpec( channels=self._out_channels[i], stride=self._out_strides[i]) for i in self.return_idx ]

[JianyuChenyyds]

可以自己手动加一下

一个是这种： @property def out_shape(self): return [ShapeSpec(channels=c) for c in self._out_channels] 一个是这种： @property def out_shape(self): return [ ShapeSpec( channels=self._out_channels[i], stride=self._out_strides[i]) for i in self.return_idx ] 可以直接加个stride=self._out_strides[i]，然后再把for循环里的self._out_channels改成len(self._out_channels)吗