train_VGG19.py error about pytorch_realtime_multi-person_pose_estimation HOT 6 CLOSED

It may caused by some path problem: https://stackoverflow.com/questions/47172770/error-215-ssize-width-0-ssize-height-0-in-function-resize

from pytorch_realtime_multi-person_pose_estimation.

You may need print if the path of that mask exist: https://github.com/tensorboy/pytorch_Realtime_Multi-Person_Pose_Estimation/blob/master/training/datasets/coco_data/COCO_data_pipeline.py#L226

from pytorch_realtime_multi-person_pose_estimation.

print(self.root + 'mask2014/val2014_mask_miss_' + img_idx + 'png')

printed ./training/dataset/COCO/images/mask2014/val2014_mask_miss_000000004947.png

ls ./training/dataset/COCO/images/mask2014/val2014_mask_miss_000000113970.png
ls: cannot access './training/dataset/COCO/images/mask2014/val2014_mask_miss_000000113970.png': No such file or directory

Filenames may be different?
mask2014 position is correct.
In mask2014, images are jpg.

from pytorch_realtime_multi-person_pose_estimation.

After change
self.mask_dir + 'mask2014/val2014_mask_miss_' + img_idx + 'png', 0)
to
self.mask_dir + 'mask2014/val2014/mask_COCO_val2014_' + img_idx + 'jpg', 0),

Error codes changed.

Loading dataset...
train dataset len: 121522
val dataset len: 4873
Bulding VGG19
load imagenet pretrained model: ./network/weight/
./training/dataset/COCO/images/mask2014/train2014/mask_COCO_train2014_000000010094.jpg
...
Traceback (most recent call last):
File "train_VGG19.py", line 310, in
train_loss = train(train_data, model, optimizer, epoch)
File "train_VGG19.py", line 157, in train
for i, (img, heatmap_target, heat_mask, paf_target, paf_mask) in enumerate(train_loader):
File "anaconda3/envs/py35/lib/python3.5/site-packages/torch/utils/data/dataloader.py", line 286, in next
return self._process_next_batch(batch)
File "anaconda3/envs/py35/lib/python3.5/site-packages/torch/utils/data/dataloader.py", line 307, in _process_next_batch
raise batch.exc_type(batch.exc_msg)
KeyError: 'Traceback (most recent call last):\n File "anaconda3/envs/py35/lib/python3.5/site-packages/torch/utils/data/dataloader.py", line 57, in _worker_loop\n samples = collate_fn([dataset[i] for i in batch_indices])\n File "anaconda3/envs/py35/lib/python3.5/site-packages/torch/utils/data/dataloader.py", line 57, in \n samples = collate_fn([dataset[i] for i in batch_indices])\n File "pytorch_Realtime_Multi-Person_Pose_Estimation/training/datasets/coco_data/COCO_data_pipeline.py", line 255, in getitem\n meta_data, mask_miss)\n File "pytorch_Realtime_Multi-Person_Pose_Estimation/training/datasets/coco_data/COCO_data_pipeline.py", line 214, in get_ground_truth\n count=count, params_transform=self.params_transform)\n File "pytorch_Realtime_Multi-Person_Pose_Estimation/training/datasets/coco_data/paf.py", line 27, in putVecMaps\n thre = params_transform['limb_width'] # limb width\nKeyError: 'limb_width'\n'

from pytorch_realtime_multi-person_pose_estimation.

It worked!

You forgot to define "params_transform['limb_width'] = 1.289" in train_VGG19.py

Thanks for your help!

from pytorch_realtime_multi-person_pose_estimation.

I remember it is 1 for VGG19 and 1.289 for stackedhourglass，that number should be different based on the size of output.
I used this method to test best sigma and limb_width for heatmap and paf:
Using different sigma and limb_width to generate ground truth heatmap and paf to see which value has the highest mAP on validation set.

from pytorch_realtime_multi-person_pose_estimation.