TypeError: forward() missing 1 required positional argument: 'img_metas'

[wxmchgithub]

I am using mmdetection version 2.28.2. I am currently using the Mask2Former model. When I use the trained model to calculate flops, params (using the thop library), or visualize using gradcam, I get an error. I don't know why this error occurs. I also get this error when using tools/get_flops.py. I hope an expert can answer this question.

[King4819]

I have the same issue. I want to calculate FLOPs for model.

[AZong76]

I have the same issue.

[AGreenApple2002]

I have the same issue,I want to count FLOPs with my own models.

[wxmchgithub]

I have the same issue,I want to count FLOPs with my own models.

I found that the reason may be that the parameter img_metas is required for forward inference of the model. There are examples of preparing img_metas in mmdet/apis/inference.py in mmdetection framework. It can be analogous to preparing forward inference data according to data preparation in inference.py. my mmdet version is 2.28.2

[King4819]

@wxmchgithub suppose I want to inference an image, how do I prepare data[img_metas] ? Thanks!!!

[wxmchgithub]

suppose I want to inference an image, how do I prepare data[img_metas] ? Thanks!!!

If you only want to perform inference on the image, you don't need to prepare img_metas yourself. demo/image_demo.py demonstrates how to perform inference on images.

[King4819]

@wxmchgithub I want to inference an image then calculate the FLOPs of model based on it.

[wxmchgithub]

@wxmchgithub I want to inference an image then calculate the FLOPs of model based on it.

@wxmchgithub I want to inference an image then calculate the FLOPs of model based on it.

You can refer to inference_detector under inference.py. It contains the preparation of img_metas.