mmpose
mmpose copied to clipboard
Published
20 hours ago •
open-mmlab
open-mmlab
Training error yolox model on custom dataset
Prerequisite
- [X] I have searched Issues and Discussions but cannot get the expected help.
- [X] The bug has not been fixed in the latest version(https://github.com/open-mmlab/mmpose).
Environment
I am trying to train yolox model on custom dataset which is annotated using cvat tool and exported to coco format.
then i followed the the MMPose_Tutorial.ipynb notebook and created the similar data loader class as TinyCoco.
i am getting the below error while training.
---------------------------------------------------------------------------
AttributeError Traceback (most recent call last)
Cell In[6], line 11
8 runner = Runner.from_cfg(cfg)
10 # start training
---> 11 runner.train()
File [~/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/mmengine/runner/runner.py:1777](http://localhost:8989/home/hcladmin/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/mmengine/runner/runner.py#line=1776), in Runner.train(self)
1773 # Maybe compile the model according to options in self.cfg.compile
1774 # This must be called **AFTER** model has been wrapped.
1775 self._maybe_compile('train_step')
-> 1777 model = self.train_loop.run() # type: ignore
1778 self.call_hook('after_run')
1779 return model
File [~/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/mmengine/runner/loops.py:96](http://localhost:8989/home/hcladmin/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/mmengine/runner/loops.py#line=95), in EpochBasedTrainLoop.run(self)
93 self.runner.call_hook('before_train')
95 while self._epoch < self._max_epochs and not self.stop_training:
---> 96 self.run_epoch()
98 self._decide_current_val_interval()
99 if (self.runner.val_loop is not None
100 and self._epoch >= self.val_begin
101 and (self._epoch % self.val_interval == 0
102 or self._epoch == self._max_epochs)):
File [~/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/mmengine/runner/loops.py:113](http://localhost:8989/home/hcladmin/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/mmengine/runner/loops.py#line=112), in EpochBasedTrainLoop.run_epoch(self)
111 self.runner.model.train()
112 for idx, data_batch in enumerate(self.dataloader):
--> 113 self.run_iter(idx, data_batch)
115 self.runner.call_hook('after_train_epoch')
116 self._epoch += 1
File [~/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/mmengine/runner/loops.py:129](http://localhost:8989/home/hcladmin/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/mmengine/runner/loops.py#line=128), in EpochBasedTrainLoop.run_iter(self, idx, data_batch)
124 self.runner.call_hook(
125 'before_train_iter', batch_idx=idx, data_batch=data_batch)
126 # Enable gradient accumulation mode and avoid unnecessary gradient
127 # synchronization during gradient accumulation process.
128 # outputs should be a dict of loss.
--> 129 outputs = self.runner.model.train_step(
130 data_batch, optim_wrapper=self.runner.optim_wrapper)
132 self.runner.call_hook(
133 'after_train_iter',
134 batch_idx=idx,
135 data_batch=data_batch,
136 outputs=outputs)
137 self._iter += 1
File [~/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/mmengine/model/base_model/base_model.py:114](http://localhost:8989/home/hcladmin/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/mmengine/model/base_model/base_model.py#line=113), in BaseModel.train_step(self, data, optim_wrapper)
112 with optim_wrapper.optim_context(self):
113 data = self.data_preprocessor(data, True)
--> 114 losses = self._run_forward(data, mode='loss') # type: ignore
115 parsed_losses, log_vars = self.parse_losses(losses) # type: ignore
116 optim_wrapper.update_params(parsed_losses)
File [~/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/mmengine/model/base_model/base_model.py:361](http://localhost:8989/home/hcladmin/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/mmengine/model/base_model/base_model.py#line=360), in BaseModel._run_forward(self, data, mode)
351 """Unpacks data for :meth:`forward`
352
353 Args:
(...)
358 dict or list: Results of training or testing mode.
359 """
360 if isinstance(data, dict):
--> 361 results = self(**data, mode=mode)
362 elif isinstance(data, (list, tuple)):
363 results = self(*data, mode=mode)
File [~/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/torch/nn/modules/module.py:1518](http://localhost:8989/home/hcladmin/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/torch/nn/modules/module.py#line=1517), in Module._wrapped_call_impl(self, *args, **kwargs)
1516 return self._compiled_call_impl(*args, **kwargs) # type: ignore[misc]
1517 else:
-> 1518 return self._call_impl(*args, **kwargs)
File [~/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/torch/nn/modules/module.py:1527](http://localhost:8989/home/hcladmin/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/torch/nn/modules/module.py#line=1526), in Module._call_impl(self, *args, **kwargs)
1522 # If we don't have any hooks, we want to skip the rest of the logic in
1523 # this function, and just call forward.
1524 if not (self._backward_hooks or self._backward_pre_hooks or self._forward_hooks or self._forward_pre_hooks
1525 or _global_backward_pre_hooks or _global_backward_hooks
1526 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1527 return forward_call(*args, **kwargs)
1529 try:
1530 result = None
File [~/mmpose/mmpose/models/pose_estimators/base.py:155](http://localhost:8989/lab/tree/demo/mmpose/models/pose_estimators/base.py#line=154), in BasePoseEstimator.forward(self, inputs, data_samples, mode)
153 inputs = torch.stack(inputs)
154 if mode == 'loss':
--> 155 return self.loss(inputs, data_samples)
156 elif mode == 'predict':
157 # use customed metainfo to override the default metainfo
158 if self.metainfo is not None:
File [~/mmpose/mmpose/models/pose_estimators/bottomup.py:70](http://localhost:8989/lab/tree/demo/mmpose/models/pose_estimators/bottomup.py#line=69), in BottomupPoseEstimator.loss(self, inputs, data_samples)
66 losses = dict()
68 if self.with_head:
69 losses.update(
---> 70 self.head.loss(feats, data_samples, train_cfg=self.train_cfg))
72 return losses
File [~/mmpose/mmpose/models/heads/hybrid_heads/yoloxpose_head.py:319](http://localhost:8989/lab/tree/demo/mmpose/models/heads/hybrid_heads/yoloxpose_head.py#line=318), in YOLOXPoseHead.loss(self, feats, batch_data_samples, train_cfg)
314 flatten_kpt_decoded = self.decode_kpt_reg(flatten_kpt_offsets,
315 flatten_priors[..., :2],
316 flatten_priors[..., -1])
318 # 2. generate targets
--> 319 targets = self._get_targets(flatten_priors,
320 flatten_cls_scores.detach(),
321 flatten_objectness.detach(),
322 flatten_bbox_decoded.detach(),
323 flatten_kpt_decoded.detach(),
324 flatten_kpt_vis.detach(),
325 batch_data_samples)
326 pos_masks, cls_targets, obj_targets, obj_weights, \
327 bbox_targets, bbox_aux_targets, kpt_targets, kpt_aux_targets, \
328 vis_targets, vis_weights, pos_areas, pos_priors, group_indices, \
329 num_fg_imgs = targets
331 num_pos = torch.tensor(
332 sum(num_fg_imgs),
333 dtype=torch.float,
334 device=flatten_cls_scores.device)
File [~/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/torch/utils/_contextlib.py:115](http://localhost:8989/home/hcladmin/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/torch/utils/_contextlib.py#line=114), in context_decorator.<locals>.decorate_context(*args, **kwargs)
112 @functools.wraps(func)
113 def decorate_context(*args, **kwargs):
114 with ctx_factory():
--> 115 return func(*args, **kwargs)
File [~/mmpose/mmpose/models/heads/hybrid_heads/yoloxpose_head.py:410](http://localhost:8989/lab/tree/demo/mmpose/models/heads/hybrid_heads/yoloxpose_head.py#line=409), in YOLOXPoseHead._get_targets(self, priors, batch_cls_scores, batch_objectness, batch_decoded_bboxes, batch_decoded_kpts, batch_kpt_vis, batch_data_samples)
408 targets_each = []
409 for i in range(num_imgs):
--> 410 target = self._get_targets_single(priors, batch_cls_scores[i],
411 batch_objectness[i],
412 batch_decoded_bboxes[i],
413 batch_decoded_kpts[i],
414 batch_kpt_vis[i],
415 batch_data_samples[i])
416 targets_each.append(target)
418 targets = list(zip(*targets_each))
File [~/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/torch/utils/_contextlib.py:115](http://localhost:8989/home/hcladmin/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/torch/utils/_contextlib.py#line=114), in context_decorator.<locals>.decorate_context(*args, **kwargs)
112 @functools.wraps(func)
113 def decorate_context(*args, **kwargs):
114 with ctx_factory():
--> 115 return func(*args, **kwargs)
File [~/mmpose/mmpose/models/heads/hybrid_heads/yoloxpose_head.py:532](http://localhost:8989/lab/tree/demo/mmpose/models/heads/hybrid_heads/yoloxpose_head.py#line=531), in YOLOXPoseHead._get_targets_single(self, priors, cls_scores, objectness, decoded_bboxes, decoded_kpts, kpt_vis, data_sample)
524 scores = cls_scores * objectness
525 pred_instances = InstanceData(
526 bboxes=decoded_bboxes,
527 scores=scores.sqrt_(),
(...)
530 keypoints_visible=kpt_vis,
531 )
--> 532 assign_result = self.assigner.assign(
533 pred_instances=pred_instances, gt_instances=gt_instances)
535 # sampling
536 pos_inds = torch.nonzero(
537 assign_result['gt_inds'] > 0, as_tuple=False).squeeze(-1).unique()
File [~/mmpose/mmpose/models/task_modules/assigners/sim_ota_assigner.py:85](http://localhost:8989/lab/tree/demo/mmpose/models/task_modules/assigners/sim_ota_assigner.py#line=84), in SimOTAAssigner.assign(self, pred_instances, gt_instances, **kwargs)
67 """Assign gt to priors using SimOTA.
68
69 Args:
(...)
82 max iou overlaps, assigned labels, etc.
83 """
84 gt_bboxes = gt_instances.bboxes
---> 85 gt_labels = gt_instances.labels
86 gt_keypoints = gt_instances.keypoints
87 gt_keypoints_visible = gt_instances.keypoints_visible
AttributeError: 'InstanceData' object has no attribute 'labels'
Reproduces the problem - code sample
from mmengine.config import Config, DictAction
from mmengine.runner import Runner
cfg.model.setdefault('data_preprocessor', cfg.get('preprocess_cfg', {}))
runner = Runner.from_cfg(cfg)
runner.train()
Reproduces the problem - command or script
from mmengine.config import Config, DictAction
from mmengine.runner import Runner
cfg.model.setdefault('data_preprocessor', cfg.get('preprocess_cfg', {}))
runner = Runner.from_cfg(cfg)
runner.train()
Reproduces the problem - error message
---------------------------------------------------------------------------
AttributeError Traceback (most recent call last)
Cell In[6], line 11
8 runner = Runner.from_cfg(cfg)
10 # start training
---> 11 runner.train()
File [~/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/mmengine/runner/runner.py:1777](http://localhost:8989/home/hcladmin/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/mmengine/runner/runner.py#line=1776), in Runner.train(self)
1773 # Maybe compile the model according to options in self.cfg.compile
1774 # This must be called **AFTER** model has been wrapped.
1775 self._maybe_compile('train_step')
-> 1777 model = self.train_loop.run() # type: ignore
1778 self.call_hook('after_run')
1779 return model
File [~/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/mmengine/runner/loops.py:96](http://localhost:8989/home/hcladmin/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/mmengine/runner/loops.py#line=95), in EpochBasedTrainLoop.run(self)
93 self.runner.call_hook('before_train')
95 while self._epoch < self._max_epochs and not self.stop_training:
---> 96 self.run_epoch()
98 self._decide_current_val_interval()
99 if (self.runner.val_loop is not None
100 and self._epoch >= self.val_begin
101 and (self._epoch % self.val_interval == 0
102 or self._epoch == self._max_epochs)):
File [~/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/mmengine/runner/loops.py:113](http://localhost:8989/home/hcladmin/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/mmengine/runner/loops.py#line=112), in EpochBasedTrainLoop.run_epoch(self)
111 self.runner.model.train()
112 for idx, data_batch in enumerate(self.dataloader):
--> 113 self.run_iter(idx, data_batch)
115 self.runner.call_hook('after_train_epoch')
116 self._epoch += 1
File [~/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/mmengine/runner/loops.py:129](http://localhost:8989/home/hcladmin/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/mmengine/runner/loops.py#line=128), in EpochBasedTrainLoop.run_iter(self, idx, data_batch)
124 self.runner.call_hook(
125 'before_train_iter', batch_idx=idx, data_batch=data_batch)
126 # Enable gradient accumulation mode and avoid unnecessary gradient
127 # synchronization during gradient accumulation process.
128 # outputs should be a dict of loss.
--> 129 outputs = self.runner.model.train_step(
130 data_batch, optim_wrapper=self.runner.optim_wrapper)
132 self.runner.call_hook(
133 'after_train_iter',
134 batch_idx=idx,
135 data_batch=data_batch,
136 outputs=outputs)
137 self._iter += 1
File [~/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/mmengine/model/base_model/base_model.py:114](http://localhost:8989/home/hcladmin/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/mmengine/model/base_model/base_model.py#line=113), in BaseModel.train_step(self, data, optim_wrapper)
112 with optim_wrapper.optim_context(self):
113 data = self.data_preprocessor(data, True)
--> 114 losses = self._run_forward(data, mode='loss') # type: ignore
115 parsed_losses, log_vars = self.parse_losses(losses) # type: ignore
116 optim_wrapper.update_params(parsed_losses)
File [~/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/mmengine/model/base_model/base_model.py:361](http://localhost:8989/home/hcladmin/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/mmengine/model/base_model/base_model.py#line=360), in BaseModel._run_forward(self, data, mode)
351 """Unpacks data for :meth:`forward`
352
353 Args:
(...)
358 dict or list: Results of training or testing mode.
359 """
360 if isinstance(data, dict):
--> 361 results = self(**data, mode=mode)
362 elif isinstance(data, (list, tuple)):
363 results = self(*data, mode=mode)
File [~/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/torch/nn/modules/module.py:1518](http://localhost:8989/home/hcladmin/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/torch/nn/modules/module.py#line=1517), in Module._wrapped_call_impl(self, *args, **kwargs)
1516 return self._compiled_call_impl(*args, **kwargs) # type: ignore[misc]
1517 else:
-> 1518 return self._call_impl(*args, **kwargs)
File [~/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/torch/nn/modules/module.py:1527](http://localhost:8989/home/hcladmin/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/torch/nn/modules/module.py#line=1526), in Module._call_impl(self, *args, **kwargs)
1522 # If we don't have any hooks, we want to skip the rest of the logic in
1523 # this function, and just call forward.
1524 if not (self._backward_hooks or self._backward_pre_hooks or self._forward_hooks or self._forward_pre_hooks
1525 or _global_backward_pre_hooks or _global_backward_hooks
1526 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1527 return forward_call(*args, **kwargs)
1529 try:
1530 result = None
File [~/mmpose/mmpose/models/pose_estimators/base.py:155](http://localhost:8989/lab/tree/demo/mmpose/models/pose_estimators/base.py#line=154), in BasePoseEstimator.forward(self, inputs, data_samples, mode)
153 inputs = torch.stack(inputs)
154 if mode == 'loss':
--> 155 return self.loss(inputs, data_samples)
156 elif mode == 'predict':
157 # use customed metainfo to override the default metainfo
158 if self.metainfo is not None:
File [~/mmpose/mmpose/models/pose_estimators/bottomup.py:70](http://localhost:8989/lab/tree/demo/mmpose/models/pose_estimators/bottomup.py#line=69), in BottomupPoseEstimator.loss(self, inputs, data_samples)
66 losses = dict()
68 if self.with_head:
69 losses.update(
---> 70 self.head.loss(feats, data_samples, train_cfg=self.train_cfg))
72 return losses
File [~/mmpose/mmpose/models/heads/hybrid_heads/yoloxpose_head.py:319](http://localhost:8989/lab/tree/demo/mmpose/models/heads/hybrid_heads/yoloxpose_head.py#line=318), in YOLOXPoseHead.loss(self, feats, batch_data_samples, train_cfg)
314 flatten_kpt_decoded = self.decode_kpt_reg(flatten_kpt_offsets,
315 flatten_priors[..., :2],
316 flatten_priors[..., -1])
318 # 2. generate targets
--> 319 targets = self._get_targets(flatten_priors,
320 flatten_cls_scores.detach(),
321 flatten_objectness.detach(),
322 flatten_bbox_decoded.detach(),
323 flatten_kpt_decoded.detach(),
324 flatten_kpt_vis.detach(),
325 batch_data_samples)
326 pos_masks, cls_targets, obj_targets, obj_weights, \
327 bbox_targets, bbox_aux_targets, kpt_targets, kpt_aux_targets, \
328 vis_targets, vis_weights, pos_areas, pos_priors, group_indices, \
329 num_fg_imgs = targets
331 num_pos = torch.tensor(
332 sum(num_fg_imgs),
333 dtype=torch.float,
334 device=flatten_cls_scores.device)
File [~/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/torch/utils/_contextlib.py:115](http://localhost:8989/home/hcladmin/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/torch/utils/_contextlib.py#line=114), in context_decorator.<locals>.decorate_context(*args, **kwargs)
112 @functools.wraps(func)
113 def decorate_context(*args, **kwargs):
114 with ctx_factory():
--> 115 return func(*args, **kwargs)
File [~/mmpose/mmpose/models/heads/hybrid_heads/yoloxpose_head.py:410](http://localhost:8989/lab/tree/demo/mmpose/models/heads/hybrid_heads/yoloxpose_head.py#line=409), in YOLOXPoseHead._get_targets(self, priors, batch_cls_scores, batch_objectness, batch_decoded_bboxes, batch_decoded_kpts, batch_kpt_vis, batch_data_samples)
408 targets_each = []
409 for i in range(num_imgs):
--> 410 target = self._get_targets_single(priors, batch_cls_scores[i],
411 batch_objectness[i],
412 batch_decoded_bboxes[i],
413 batch_decoded_kpts[i],
414 batch_kpt_vis[i],
415 batch_data_samples[i])
416 targets_each.append(target)
418 targets = list(zip(*targets_each))
File [~/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/torch/utils/_contextlib.py:115](http://localhost:8989/home/hcladmin/anaconda3/envs/mmpose-env/lib/python3.8/site-packages/torch/utils/_contextlib.py#line=114), in context_decorator.<locals>.decorate_context(*args, **kwargs)
112 @functools.wraps(func)
113 def decorate_context(*args, **kwargs):
114 with ctx_factory():
--> 115 return func(*args, **kwargs)
File [~/mmpose/mmpose/models/heads/hybrid_heads/yoloxpose_head.py:532](http://localhost:8989/lab/tree/demo/mmpose/models/heads/hybrid_heads/yoloxpose_head.py#line=531), in YOLOXPoseHead._get_targets_single(self, priors, cls_scores, objectness, decoded_bboxes, decoded_kpts, kpt_vis, data_sample)
524 scores = cls_scores * objectness
525 pred_instances = InstanceData(
526 bboxes=decoded_bboxes,
527 scores=scores.sqrt_(),
(...)
530 keypoints_visible=kpt_vis,
531 )
--> 532 assign_result = self.assigner.assign(
533 pred_instances=pred_instances, gt_instances=gt_instances)
535 # sampling
536 pos_inds = torch.nonzero(
537 assign_result['gt_inds'] > 0, as_tuple=False).squeeze(-1).unique()
File [~/mmpose/mmpose/models/task_modules/assigners/sim_ota_assigner.py:85](http://localhost:8989/lab/tree/demo/mmpose/models/task_modules/assigners/sim_ota_assigner.py#line=84), in SimOTAAssigner.assign(self, pred_instances, gt_instances, **kwargs)
67 """Assign gt to priors using SimOTA.
68
69 Args:
(...)
82 max iou overlaps, assigned labels, etc.
83 """
84 gt_bboxes = gt_instances.bboxes
---> 85 gt_labels = gt_instances.labels
86 gt_keypoints = gt_instances.keypoints
87 gt_keypoints_visible = gt_instances.keypoints_visible
AttributeError: 'InstanceData' object has no attribute 'labels'
Additional information
- it should start training
