I encountered an error when working on the prompt_tuning project: TypeError:loss() missing 1 required positional argument: 'batch_data_samples'.

[JoyRiderJie]

My configuration file is derived from configs/prompt_tuning_coco/yolo_world_v2_l_vlpan_bn_2e-4_80e_8gpus_mask-refine_prompt_tuning_coco.py, however, I have modified the model loading part to set load_from='pretrained_models/l_stage1-7d280586.pth'.

The detailed error message is as follows:

Traceback (most recent call last):
  File "tools/train.py", line 120, in <module>
    main()
  File "tools/train.py", line 116, in main
    runner.train()
  File "/mnt/data3/zxj2024/workplace/YOLO-World/third_party/mmengine/mmengine/runner/runner.py", line 1777, in train
    model = self.train_loop.run()  # type: ignore
  File "/mnt/data3/zxj2024/workplace/YOLO-World/third_party/mmengine/mmengine/runner/loops.py", line 98, in run
    self.run_epoch()
  File "/mnt/data3/zxj2024/workplace/YOLO-World/third_party/mmengine/mmengine/runner/loops.py", line 115, in run_epoch
    self.run_iter(idx, data_batch)
  File "/mnt/data3/zxj2024/workplace/YOLO-World/third_party/mmengine/mmengine/runner/loops.py", line 131, in run_iter
    outputs = self.runner.model.train_step(
  File "/mnt/data3/zxj2024/workplace/YOLO-World/third_party/mmengine/mmengine/model/base_model/base_model.py", line 114, in train_step
    losses = self._run_forward(data, mode='loss')  # type: ignore
  File "/mnt/data3/zxj2024/workplace/YOLO-World/third_party/mmengine/mmengine/model/base_model/base_model.py", line 361, in _run_forward
    results = self(**data, mode=mode)
  File "/mnt/data3/zxj2024/conda_envs/yolo_world/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1110, in _call_impl
    return forward_call(*input, **kwargs)
  File "/mnt/data3/zxj2024/workplace/YOLO-World/third_party/mmdetection/mmdet/models/detectors/base.py", line 92, in forward
    return self.loss(inputs, data_samples)
  File "/mnt/data3/zxj2024/workplace/YOLO-World/yolo_world/models/detectors/yolo_world.py", line 166, in loss
    losses = self.bbox_head.loss(img_feats, txt_feats,
TypeError: loss() missing 1 required positional argument: 'batch_data_samples'

The error occurs in the SimpleYOLOWorldDetector class within the file yolo_world/models/detectors/yolo_world.py, specifically spanning lines 163 to 167.

[zhuzhanshi]

I was able to fix this by passing txt_masks into loss_by_feat().

In YOLOWorldHead.loss(), the current “Fast version” only passes outs + (bboxes_labels, img_metas), but loss_by_feat() expects batch_text_masks as a required positional argument. Since outs only contains three outputs, the arguments become misaligned and Python reports a missing batch_img_metas.

The fix is simply to insert txt_masks (which is batch_text_masks) before bboxes_labels:

outs = self(img_feats, txt_feats, txt_masks)
- loss_inputs = outs + (batch_data_samples['bboxes_labels'],
-                       batch_data_samples['img_metas'])
+ loss_inputs = outs + (txt_masks,
+                       batch_data_samples['bboxes_labels'],
+                       batch_data_samples['img_metas'])
losses = self.loss_by_feat(*loss_inputs)

After this change, training runs normally.

[inarticular]

I was able to fix this by passing txt_masks into loss_by_feat().

In YOLOWorldHead.loss(), the current “Fast version” only passes outs + (bboxes_labels, img_metas), but loss_by_feat() expects batch_text_masks as a required positional argument. Since outs only contains three outputs, the arguments become misaligned and Python reports a missing batch_img_metas.

The fix is simply to insert txt_masks (which is batch_text_masks) before bboxes_labels:

outs = self(img_feats, txt_feats, txt_masks)

• loss_inputs = outs + (batch_data_samples['bboxes_labels'],

•                   batch_data_samples['img_metas'])
  

• loss_inputs = outs + (txt_masks,

•                   batch_data_samples['bboxes_labels'],
  

•                   batch_data_samples['img_metas'])
  

losses = self.loss_by_feat(*loss_inputs)
After this change, training runs normally.

Hi, thanks for your suggestion.

I modified the code to:

loss_inputs = outs + (txt_masks, batch_data_samples['bboxes_labels'], batch_data_samples['img_metas'])

but I still get the following error:

TypeError: loss() missing 1 required positional argument: 'batch_data_samples'

If it is convenient for you, could you please share your complete yolo_world_head.py and yolo_world.py files for reference? I would really appreciate it, as it would help me compare them with my current version and better understand what else may need to be adjusted.

Thank you again for your time and help.