facebookresearch / c3dpo_nrsfm Goto Github PK

urllib3.exceptions.NewConnectionError: <urllib3.connection.HTTPConnection object at 0x7fb4322c4e10>: Failed to establish a new connection: [Errno 111] Connection refused

I am trying to train the model to make it work for 3D reconstruction for face. I used a 2D face landmark detector to detect 2D landmarks on face images. Then I trained the model using the same config file h36m.yaml except I set shape_basis_size=1000 as in your ICCV presentation talk slides. The training converges very fast, but when I test it, the result is not as good as seen from the image I pasted. Could you point me to how to make it work for 3D reconstruction for face? Thank you.

I'm trying to train a model for coco human dataset. However, there is no coco url in DATASET_URL. What should I do?

Hello, I want to train your model by myself, but I have confuse about the input kp_loc, how can it get from 2D keypoint pixel annotations, can you provide the preproduce code? please. I will very appreciate for your kindness.

I am facing this weird error, dont really know what to do about it. Was running it in a CONDA env. as instructed. Below can be seen a screenshot.

Hey, it seems that you assumed the scale of the person is known, so if we run the model on a new dataset, and don't have human scale, we can not predict absolute 3d keypoints coordinates, right?

Hi，How to get kp_loc from kp_original in h36m dataset?

when set projection_type: perspective in exconfig.yaml and run demo.py
but assert K is not None in def forward(self, kp_loc=None, kp_vis=None, class_mask=None, K=None, **kwargs):

Error info:

requests.exceptions.ConnectionError: HTTPConnectionPool(host='localhost', port=8097): Max retries exceeded with url: /events (Caused by NewConnectionError('<urllib3.connection.HTTPConnection object at 0x000002413CA92E10>: Failed to establish a new connection: [WinError 10061] 由于目标计算机积极拒绝，无法连接。',))

Hello, we are very interested in your research, but we need to download the sample file from visdom to run the program. Due to the limitation of Chinese network, visdom cannot be installed and used normally. Can you provide the static folder of visdom for us to manually replace the static file in the visdom directory. Thanks.

Hi, I run the demo.py but find the visualization results look totally different from the video of your project pages. (Your demo video is cooler! ) How can I draw the 3D pose just like the one on your project pages? Many thanks!

How do I train this model on a custom dataset? What are the steps I need to follow?

Hello, Can it be used to inferent a wild-image? And what is the keypoints' meaning? Or to say, if I get the pixel of keypoint of a wild-image, what should I do to pre-process them and get the kp_loc?

Dear All, thank you for your excellent contribution to the community. I was wondering about your very nice visualizations of the 3D wireframes: are there scripts you could share that automate the shading/ray tracing that you're using? Is this blender?

Thanks,
Tim

Hi,
I have several images of humans in .jpg format. I am trying to extract 3D poses from these images. How am I to apply c3dpo_nrsfm to my input images?

Thanks
Mritula

Hi, I notice you have normalized the original 2D keypoint coordinate to kp_loc in the dataset file, but how do you normalize it? Thanks!

Hi,
I am trying to implement c3dpo_nrsfm as in github to learn it. I am neither getting any extracted 3D points nor an error. It stays as below

I tried the following commands in the sequence

1. python -m visdom.server
2. python ./experiment.py --cfg_file ./cfgs/h36m.yaml

and

1. python -m visdom.server
2. python ./demo.py

thanks
Mritula

Hi,
How do I visualize the results on PASCAL3D+ dataset.
The models and the keypoint annotations given are with respect to 124 keypoints. However, for example for a car I wish to visualize only the valid keypoints (12 in the case of PASCAL3D+). When I try to visualize by changing the model in demo.py, it plots the invalid keypoints as well.
Would really appreciate some help!