Thank you for your hard work. Great job! I want to try your method, but I haven't got the human 3.6m dataset yet. Can I try with other datasets like the SURREAL dataset or DenseposeCoCo?

After I trained it several epoch ,The problem shows that ： Traceback (most recent call last): File "nohup_train.py", line 120, in visualizer.save_results(model.get_current_visuals(), epoch, i) File "/data/home/data_utils/visualizer.py", line 48, in save_results resampled_verts = self.UV_sampler.resample(self.tensor2numpy(fake_UV)) File "/data/home/data_utils/uv_map_generator.py", line 409, in resample vt_3d = np.stack(vt_3d) File "/opt/conda/lib/python3.6/site-packages/numpy/core/shape_base.py", line 416, in stack raise ValueError('all input arrays must have the same shape') ValueError: all input arrays must have the same shape can you help me analyze the problem?

Hi,

I am trying out this repo, and am trying to generate a 3D SMPL model out of a 2D image using the "in-the-wild" mode.

I used the sample image found in the teaser folder, and the UV Maps & SMPL Model generated appears cranky.

Not sure if I am doing something wrong... Using the below parameters:

• Using the encoder & decoder weights
• Using model_lsp.pkl generated via instructions here
• Using radvani.pickle generated from template.obj via UV_Map_Generator class, and the resulting barycentric_h0256_w0256_radvani.pickle
• Saw this issue and tried using [-1, 1] (by default via densebody_dataset.py) & [0, 255] UV Maps output, but both gave similar results above

Hi Lotayou, Thank you for your work and detailed instructions. I followed your steps to prepare Datasets and UV.

However I got wrong results on the model I trained by myself and I also tested on your pre-trained model.

Maybe My SMPL model isn't right? And I have no ideal which step I failed. Any help would be deeply appreciated! ^_^

@Lotayou Thanks for your awesome work, I want train the human3.6M dataset, but I don't know what data do I need to download. Does the code need videos, pose, segments ? or I need download them all. Hope you can help me, thanks in advace.

when I ran data_washing.py, it reported missing annotation_large.h5. How is this file generated? I use https://github.com/anibali/h36m-fetch to preprocess human3.6m it only generated anno.h5 file for each subject.

I have used the original code and the provided toy dataset and trained with enough time, but the predicted UV map I get seems wrong... Does anyone encounter this problem? How to solve it? Thanks!

This is the UV map generated during the training process...(I have trained like, about 200 epochs...)

Hello, I can not find the model.pkl when I run the smpl_torch_batch.py. Could you tell me how to find it ? Besides, there is an 'indexerror: index 6890 is out of bounds for axis 0 with size 6890' when I run the triangulation.py, I do not know if the error is caused by the model.pkl, maybe there are some other mistakes? I put the 'for (i,line) in enumerate(lines):' modify to 'for (i,line) in enumerate(lines[0:6890]):', the indexerror is lost.But there is another 'indexerror:index 13790 is out of bounds for axis 0 with size 13776.

Hi, Thanks for sharing this amazing work. This will strongly support my research. I need to develop a 3d model with the scanned image. This project has the first part of developing the skeleton. How can I add texture of the same image to this?

not an issue but a question, let's say if I have 4 cameras around the same person and can I put 4 images into the singe UV map and generate this person's better 3D model?

_min = np.amin(verts, axis=0, keepdims=True) _max = np.amax(verts, axis=0, keepdims=True) verts = (verts - _min) / (_max - _min)``

I think here is wrong, the scale of x,y,z is different

Hi Lotayou, thanks for your project and nice instructions. I just want to ask that do you have the code to process the H36M downloaded from the original website. It seems that the code is made just for the toy dataset. Thanks

Hi @Lotayou, thanks for your great effort! But I found the link to pretrained encoder and decoder cannot be reached. Would you mind reupload them again? Thanks!

I am not sure if the smpl_fbx_template.obj file I generated is correct. After I run uv_map_generator.py, I get an error. ValueError: shape mismatch: value array of shape (7521,3) could not be broadcast to indexing result of shape (7576,3)

hi, I modify the train.sh as :

python train.py --name resnet_radvani_32000_20190415 --model resnet --netD conv-up --batch_size 4 --max_dataset_size 32000 --niter 20 --niter_decay 50 --save_result_freq 250 --save_epoch_freq 2 --ndown 6 --data_root /home/liwensh/data

and run about 16 hours(44 epoch), and at the end of log.txt shows:

epoch 43 iter 6499:  l1: 11.288669 tv: 1.522304 total: 11.288669
epoch 43 iter 6749:  l1: 11.599895 tv: 0.667862 total: 11.599895
epoch 43 iter 6999:  l1: 11.125267 tv: 1.277602 total: 11.125267
epoch 43 iter 7249:  l1: 11.893361 tv: 1.366742 total: 11.893361
epoch 43 iter 7499:  l1: 11.343329 tv: 1.228081 total: 11.343329
epoch 43 iter 7749:  l1: 11.397069 tv: 1.426213 total: 11.397069
epoch 43 iter 7999:  l1: 11.519998 tv: 0.664876 total: 11.519998
epoch 44 iter 249:  l1: 11.183926 tv: 1.258252 total: 11.183926
epoch 44 iter 499:  l1: 11.555054 tv: 1.201256 total: 11.555054
epoch 44 iter 749:  l1: 12.041154 tv: 1.312884 total: 12.041154
epoch 44 iter 999:  l1: 11.605458 tv: 0.706056 total: 11.605458
epoch 44 iter 1249:  l1: 11.589639 tv: 1.093558 total: 11.589639
epoch 44 iter 1499:  l1: 11.533211 tv: 1.338729 total: 11.533211
epoch 44 iter 1749:  l1: 11.822362 tv: 1.297630 total: 11.822362
epoch 44 iter 1999:  l1: 12.410873 tv: 1.159959 total: 12.410873
epoch 44 iter 2249:  l1: 11.855060 tv: 1.531642 total: 11.855060

the total have not changed much since the 5th epoch. and the inter output is strange(44 epoch): I wonder if it it because the batch size is too small, since I have no enough GPU memory. Or maybe other option set I am wrong?

Hello, i have tried your code and make some changes(tips) to perform better. Hope it would help：

1. L1 loss weight . The L1 loss weight generated is wrong accrodding to your code and your uv-map-template after dilation. UV map shouldnot be dilated
2. TV loss. I added tv loss with corrected weights. The TV loss weights should smaller than L1 loss two order of magnitudes . The results showed it improved a little.
3. From UV map to Verts. changed SAMPLE algorithm little.
4. Increase joints-center verts loss weight.

However，the MPJPE-PA i calculated is about 62, much lager than the paper listed. The MPJPE-PA between joints from UV-map-generated-form-gt3d and gt3ds is about 22mm, it also much larger than paper listed.