Tf-pose-estimation: Shape error when trainning cmu model

[erwangccc]

Hi,guys,I meet this error when trainning cmu on coco2017 dataset,is there a solution for this error?
I would appreciate it if you have advice to share with me.

• OS:ubantu16.04
• Batchsize=64

[2019-02-26 11:36:49,227] [pose_dataset] [ERROR] err type2, err=Cannot feed value of shape (64, 184, 184, 19) for Tensor 'heatmap:0', which has shape '(64, 46, 46, 19)', placeholders=[<tf.Tensor 'image:0' shape=(64, 368, 368, 3) dtype=float32>, <tf.Tensor 'heatmap:0' shape=(64, 46, 46, 19) dtype=float32>, <tf.Tensor 'vectmap:0' shape=(64, 46, 46, 38) dtype=float32>]
[2019-02-26 11:36:49,230] [pose_dataset] [ERROR] Exception in Thread-1:Cannot feed value of shape (64, 184, 184, 19) for Tensor 'heatmap:0', which has shape '(64, 46, 46, 19)'

[ildoonet]

Can you provide your training command?

[erwangccc]

@ildoonet thanks for your reply,the command is as following:
python train.py --model=cmu --datapath=data/coco_data/annotations --imgpath=data/coco_data/ --batchsize=64 --lr=0.001 --modelpath=data/ --gpus=1 --logpath=data/

input width,input height are 368.

[ildoonet]

Weird. My code works fine in my machine. Can you try with the branch 'dev/arcthiecture-mobilenet-v2' ?

[erwangccc]

@ildoonet Thanks for your advice.But i was in the master branch for a long time,it may not be very convenient for me.
Is there another solution for this error?Thank you.

[ildoonet]

@erwangccc Please pull the latest version of master branch and try again.

[erwangccc]

@ildoonet Thank you very much for your reply.
I am sorry for my late reply. I will try the lastest version of master branch and try again.

[Tjay456]

I Have the exact same error trying to retrain mobilenet_thin with train.py datapath=E:\COCO\annotations --imgpath=E:\COCO\ --gpus=1 --model=mobilenet_thin --input-width=368 --input-height=368
any suggestions?