thmoa / tex2shape Goto Github PK

Can we expect python 3 support?

From what I saw the requirements doens't work either? Chumpy (latest) is not support on python 3?

Python 2.7 is EoL it doesn't make sense to keep using it?

Workaround for Chumpy:
pip install git+https://github.com/scottandrews/chumpy.git@fe51783e0364bf1e9b705541e7d77f894dd2b1ac

if sys.version_info[0] == 3: import _pickle as pkl else: import cPickle as pkl

This leads to the error:

return codecs.charmap_decode(input,self.errors,decoding_table)[0] UnicodeDecodeError: 'charmap' codec can't decode byte 0x9d in position 350: character maps to <undefined>

How to register the SMPL model against your scans

Hi~ Thanks for your great work!
Is there any plan to share the training code? Thanks!

Hi, could you share the code for generating UV textures from HMR output as shown in Figure 7 in the paper? I couldn't find online how to do this. Thank you

Hi, thanks for your wonderful work!

I would like to ask how did you get the dp_uv_lookup_256.npy file?
Look forward to your reply.

Thank you!

Not an issue. Just a comment about opencv installation. Simply using pip install opencv leads to errors. That is because python2.7 support was dropped from opencv=4.3.0.36 onwards.
Using pip install opencv-python==4.2.0.32 worked for me.

Other errors I faced and how I solved them:
1] ERROR:

Could not load dynamic library 'libnvinfer.so.6.
Cannot dlopen some GPU libraries. 
Please make sure the missing libraries mentioned above are installed properly if you would like to use GPU.

SOLUTION:
pip install tensorflow==1.13.1

2] ERROR:

Traceback (most recent call last):
  File "run.py", line 12, in <module>
    from models.tex2shape_model import Tex2ShapeModel
  File "/mnt/t-shingane/tex2shape/models/tex2shape_model.py", line 1, in <module>
    from keras.models import Model
  File "/mnt/envs/tex2shape/lib/python2.7/site-packages/keras/__init__.py", line 22, in <module>
    from keras import distribute
  File "/mnt/envs/tex2shape/lib/python2.7/site-packages/keras/distribute/__init__.py", line 18, in <module>
    from keras.distribute import sidecar_evaluator
  File "/mnt/envs/tex2shape/lib/python2.7/site-packages/keras/distribute/sidecar_evaluator.py", line 180
    f'No checkpoints appear to be found after {_CHECKPOINT_TIMEOUT_SEC} '
                                                                        ^
SyntaxError: invalid syntax

SOLUTION:
pip install keras==2.3.0 (I had initially just run pip install keras (as mentioned in README) which installed keras 2.9.0 and was giving the aforementioned error)

How can we get the posed mesh ? In your implementation, mesh is always in template pose.

HI, thanks for the exciting job. I tried to extract an UV map of an example image based on DensePose, the output is slightly different with the corresponding example UV map, but the resulting 3D model is much worse than that of examples. Could you please share some details of the DensePose ? (for example, the backbone models listed in the Model Zoos) Or, did i miss something to get this bad result? @thmoa Hope to get a reply, thanks.

Hi thmoa,
Thanks for your great program that helps me a lot.
I have noticed that there were some canonical UV maps as templates used in a series of your papers, such as,

[1] Tex2Shape: Detailed Full Human Body Geometry From a Single Image
[2] Video Based Reconstruction of 3D People Models
[3] Detailed Human Avatars from Monocular Video

where the UV maps look like

I guess all of them are the same UV maps since they have similar shapes and body-partitions.

I want to know what methods/tools/softwares did you used to generate this UV map. Did you use Blender/Maya?

Thanks~

Great work! Thanks for providing a lot of interesting insights. I'm trying to train my own model and encounter some unclear parts for generating the ground truth displacement maps. Given your detailed registration of the SMPL to the scans, you "simply render detailed UV displacement and normal maps". Those "encode the free-form offsets, that are not part of SMPL".

Does this mean you render uv maps from the default SMPL model, then your detailed registered SMPL model and then take the difference between those (called free-form offsets in your paper)?

If my understanding of this is wrong, I would be very happy if you could provide some further information of how to reproduce uv displacements maps encoding free-form offsets to SMPL. Thank you!

hi thmoa:
thanks for your greate program. this is very effient. but now I meet a problem, Can you tell me how get a texture for this model ?
can I use this <semantic_human_texture_stitching >?? thanks for your help

Hi,
Amazing work.
Does the code contain method to generate Texture for the 3D model generated , if not can you please give some heads up on how to create the 3d texture for the OBJ file.
Thanks

The paper takes advantage of 'rich illumination and shading information contained in RGB values'. But will the predicted result be related to color? (For example, the network can rebuild a person in white clothes but can't handle people who wear black clothes well) If so, how does the network solve it?

Thanks!

Hi, @thmoa , thanks a lot for providing this excellent work!
I installed and made it work with python3.7.
Thus, which software can I use to see the result?

Hi,
why do the results of different input images in the same pose?

Is there any chance to get updated repo ? The requirements file are really old.

Thanks.

hello, thmoa:
thanks for your program!! when i look around this paper ,I find the different pose in the experiment part, Can you how can I get the result with poses? I run the pragram with no pose before. looking for your reply，thanks

the bellow is my result

Hello, thank you for you work, can you provide the requestment for runing the demo?
i use python3,but meet some error...