search

ViTAE-Transformer / ViTPose

The official repo for [NeurIPS'22] "ViTPose: Simple Vision Transformer Baselines for Human Pose Estimation" and [TPAMI'23] "ViTPose++: Vision Transformer for Generic Body Pose Estimation"
Apache License 2.0
1.37k stars 186 forks source link

Config file Use: model and loaded state dict do not match exactly #82

Open Lingyun97 opened 1 year ago

Lingyun97 commented 1 year ago

I am using the configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/ViTPose_base_coco_256x192.py to load split Vitbase+ checkpoint. The performance is great, but there's warning:The model and loaded state dict do not match exactly image

or did I used the wrong config? I'm not sure how to use the vitPose+_base_coco+aic+mpii+ap10k+apt36k+wholebody_256x192_udp.py, it output the error "KeyError: 'dataset_idx'" image

jun297 commented 1 year ago

Hi, did you solve the problem? @Lingyun97 I have the similar issue. unexpected key in source state_dict: associate_keypoint_heads.0.deconv_layers.0.weight, associate_keypoint_heads.0.deconv_layers.1.weight, associate_keypoint_heads.0.deconv_layers.1.bias, associate_keypoint_heads.0.deconv_layers.1.running_mean, associate_keypoint_heads.0.deconv_layers.1.running_var, associate_keypoint_heads.0.deconv_layers.1.num_batches_tracked, associate_keypoint_heads.0.deconv_layers.3.weight, associate_keypoint_heads.0.deconv_layers.4.weight, associate_keypoint_heads.0.deconv_layers.4.bias, associate_keypoint_heads.0.deconv_layers.4.running_mean, associate_keypoint_heads.0.deconv_layers.4.running_var, associate_keypoint_heads.0.deconv_layers.4.num_batches_tracked, associate_keypoint_heads.0.final_layer.weight, associate_keypoint_heads.0.final_layer.bias, associate_keypoint_heads.1.deconv_layers.0.weight, associate_keypoint_heads.1.deconv_layers.1.weight, associate_keypoint_heads.1.deconv_layers.1.bias, associate_keypoint_heads.1.deconv_layers.1.running_mean, associate_keypoint_heads.1.deconv_layers.1.running_var, associate_keypoint_heads.1.deconv_layers.1.num_batches_tracked, associate_keypoint_heads.1.deconv_layers.3.weight, associate_keypoint_heads.1.deconv_layers.4.weight, associate_keypoint_heads.1.deconv_layers.4.bias, associate_keypoint_heads.1.deconv_layers.4.running_mean, associate_keypoint_heads.1.deconv_layers.4.running_var, associate_keypoint_heads.1.deconv_layers.4.num_batches_tracked, associate_keypoint_heads.1.final_layer.weight, associate_keypoint_heads.1.final_layer.bias, backbone.cls_token

bilguudeiblgd commented 5 months ago

@jun297 May be too late but it seems like it's because of the config. Vitpose+ has associate_keypoint_heads which is the decoders for different representations(COCO, MPII, etc...), but the configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/ViTPose_base_coco_256x192.py doesn't have those. Either you can manually try including those or configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/vitPose+_base_coco+aic+mpii+ap10k+apt36k+wholebody_256x192_udp.py just use this.

hughkhu commented 5 months ago

@jun297 May be too late but it seems like it's because of the config. Vitpose+ has associate_keypoint_heads which is the decoders for different representations(COCO, MPII, etc...), but the configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/ViTPose_base_coco_256x192.py doesn't have those. Either you can manually try including those or configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/vitPose+_base_coco+aic+mpii+ap10k+apt36k+wholebody_256x192_udp.py just use this.

Same problem. Also "KeyError: 'dataset_idx'" error occurs when I try to use configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/vitPose+_base_coco+aic+mpii+ap10k+apt36k+wholebody_256x192_udp.py.