Official implementation of the ICCV 2021 paper: "The Power of Points for Modeling Humans in Clothing".

https://github.com/qianlim/POP/blob/c7e045e6a701de9581ffd7cfab67475f4eb979f0/lib/utils_io.py#L35 Hi, I found that the above code seems to generate transposed UV coordinates, i.e. the UV coordinates of the first row of the image ([height, width, channel] format) grid are [0, 0], [0, 1], [0, 2], ..., [ 0, H-1].

Hi,

Thanks for sharing your great work! I notice that the scale and translation of "scan_pc" and "body_verts" in the provided packed data are different from meshes in the original CAPE dataset. What is the scale and translation parameters you used to normalize the meshes?

Thanks!

(POP) siduo@siduo-NK50S5-SZ:~/下载/POP-main$ python main.py --config configs/config_demo.yaml --mode test

------------------------Loading checkpoint POP_pretrained_ReSynthdata_12outfits_epoch00400_model.pt

------------------------Test model with checkpoint at epoch 400

------------------------Eval on test data, seen outfits, unseen poses... Loading test data... Loading rp_alexandra_posed_006, 0 examples.. 0it [00:00, ?it/s] Data loaded, in total 0 test examples.

Evaluating... 0it [00:00, ?it/s] Traceback (most recent call last): File "main.py", line 354, in main() File "main.py", line 256, in main test_stats = test_seen_clo( File "/home/siduo/下载/POP-main/lib/infer.py", line 316, in test_seen_clo test_m2s /= n_test_samples ZeroDivisionError: division by zero

For evaluation during training, the value of the "geom_featmap" seems to be reset to the most recently saved value. https://github.com/qianlim/POP/blob/a05404dd44173809abfee4e2ebef57bbc524cfde/main.py#L193-L195 The "geom_featmap" is saved every 50 epochs, while the default value of the "args.val_every" is set to 20. Therefore, the value of the "geom_featmap" will be reset several times. For example, at the 20th or 40th epoch, the "geom_featmap" will be set to the initial value of the 0th epoch. Is this a bug?

Hi. You don't seem to be using the chamfer distance directly as the loss, would the performance be better? https://github.com/qianlim/POP/blob/a05404dd44173809abfee4e2ebef57bbc524cfde/lib/train.py#L87 https://github.com/qianlim/POP/blob/a05404dd44173809abfee4e2ebef57bbc524cfde/lib/train.py#L98

Hi,

In Fig 4. of the paper, you mentioned that the point clouds are rendered with surfels. Could you kindly share the code for the point rendering with surfels?

Thanks!

Hi, thanks for sharing this great work, I'm wondering how can i get the original meshes from which the scan point cloud is generated? Are they included somewhere in the dataset we downloaded or they are not provided? Thanks!!

Hi. I am confused about the transformation matrix T. The ri in Eq(2) is the displacement of the point on posed human body without outfits, why not add pi directly? That is: xi=ri+pi. Could you explain the T and how to calculate it?

Hello, it seems in the ReSynth Dataset download page, the below texts are without hyperlinks, so I cannot download demo data.

Download links:

Data_for_POP_demo: packed npz files | high-res point clouds

Hi, thank you for sharing the code. I noticed a "no distribution" term in your license. I wonder whether I can open source my own code which is partly based on POP (for noncommercial research purpose only, of course).

Hi, great work for human modeling, and I wonder to know what's mean of vtransf, I just found that: "a numpy array of shape (N_vert, 3, 3): the local-to-global transformation matrix of each vertex on the posed SMPL-X body." what's local-to-global mean and how to get it?

Thanks for sharing the code of the wonderful work!

I had tried to test the code with pre-trained weight for ReSynthedata provided in the link in Readme. With the default configuration setting, there is a dimension mismatch between the pre-trained weight and the defined network. Specifically, for unet_posefeat defined by this module, UnetNoCond7DS the layer upconvC5 should have input channel size as 256. However, in the pre-trained model, its dimension is 384 which yields dimension mismatch error during loading that model like below.

Is there any way to fix this?

python main.py --config configs/config_demo.yaml --mode test ，出现

RPly: Unable to open file Read PLY failed: unable to open file: /mnt/data2/cv610/anaconda3/envs/SCALEcode/POP/lib/../data/resynth/scans/rp_beatrice_posed_005/96_longshort_flying_eagle/00020_

------Step 1: Optimizing w.r.t. UNSEEN scan with full_gt

Traceback (most recent call last): File "main.py", line 345, in main() File "main.py", line 332, in main **model_config File "/mnt/data2/cv610/anaconda3/envs/SCALEcode/POP/lib/infer.py", line 414, in test_unseen_clo num_unseen_frames=num_unseen_frames File "/mnt/data2/cv610/anaconda3/envs/SCALEcode/POP/lib/infer.py", line 187, in reconstruct random_subsample_scan=random_subsample_scan, num_unseen_frames=num_unseen_frames) File "/mnt/data2/cv610/anaconda3/envs/SCALEcode/POP/lib/infer.py", line 103, in model_forward_and_loss lnormal, closest_target_normals = normal_loss(pred_normals, target_pc_n, idx_closest_gt) File "/mnt/data2/cv610/anaconda3/envs/SCALEcode/POP/lib/losses.py", line 21, in normal_loss target_normals_chosen = torch.gather(target_normals, dim=1, index=nearest_idx) RuntimeError: setStorage: sizes [1, 47911, 3], strides [0, 0, 1], storage offset 0, and itemsize 4 requiring a storage size of 12 are out of bounds for storage of s

Hi, how to inference on custom data? like a single static scan. Need it be fitted to a spml-x model? If I want to animate this scan, how to get a pose sequence like the "*.npz" file in you demo data? Can you please give the instruction like https://github.com/qianlim/SCALE/issues/6#issuecomment-888125275 ? Again, thanks for your excellent work!