search

facebookresearch / DensePose

A real-time approach for mapping all human pixels of 2D RGB images to a 3D surface-based model of the body
http://densepose.org
Other
6.95k stars 1.29k forks source link

It is too hard to install. #4

Open yywangvr opened 6 years ago

yywangvr commented 6 years ago

Hello, I just find the good work in a news and I try to install, but just want to say it is a little complicated to install. Why don't you make the life easier,making it like: -pip install DensePose.

:) hahaha, anyway, good work.

Tetsujinfr commented 6 years ago

Definitelly agreed. Intallwise it does not compare favourably with openpose from CMU, which is a rolls to get working and deliver very stable inferences on various test data. Here, just impossible to get this code to work (yet). Also, I am surprised by the very limited set of test results available so far. Try to google densepose, you find always the 2 or 3 same videos (incl the one from the paper). If it was so impressive and cool, shall we not see an ocean of test results 2 month after the open sourcing on git? Pls someone, tell me I am missing something here.

siamaksalman commented 6 years ago

use this guide : denseposeInstalltion

janitha09 commented 6 years ago

If you have AWS, this is a little bit easier thanks to this https://gist.github.com/matsui528/6d223d17241842c84d5882a9afa0453a

Tetsujinfr commented 6 years ago

I managed to get it installed after quite a lot of troubles. I did re-install from a clean Ubuntu install. Did install Cuda 8.0 with Cudnn 7. I did compile caffe2 from source and used miniconda for the python and other dependency stack. My caffe2 installed libraries are under usr/local/lib which I did put in the LD_LIBRARY_PATH and my caffe2 modules are in the pytorch/build folder which I did add to my PYTHONPATH. I had to fix a small issue with ATen headers which were not installed properly (this will likely be fixed with newer versions of caffe2 code), I did mention the details under issue #97. Here is a test video I processed with the FB pre-trained model: https://youtu.be/apiwISDRB3k

Quite an effort to get this code to run but happy it works fine in the end.

phpmind commented 6 years ago

Is there any real "one" script available to install this DensePose code ? Who is using this in a real word now? Any link will be useful.

jaggernaut007 commented 6 years ago

Use the docker image. Its a piece of cake to install

helenchg commented 5 years ago

Is there an easier way to install DensePose on Windows10 using Anacoda? I cant get pass Caffe2 installation. I managed to get PyTorch though.

rafikg commented 5 years ago

Who installed the code using conda environment?

sklipnoty commented 5 years ago

Would love for a clear guide on how to install on Windows. Also python3 support.

Johnqczhang commented 5 years ago

Who installed the code using conda environment?

@Gouiaa Hi, I give an installation guide using conda on Linux system.

rafikg commented 5 years ago

@Johnqczhang, I already installed this project. However, the frame rate is too low. A lot of persons had too low frame rate... So, I think it is fake project... just because it is coming from facebook

Tetsujinfr commented 5 years ago

Probably your inference is not using your gpu and defaulting on your cpu. Do you have some warning msg about missing cudnn or incorrect version when you run the model? I agree it is a pain in the neck to run properly and when finally I managed to get all run smoothly, my linux setup got corrupted though the video card driver updates I made (so it never worked again after rebooting my PC).so there is a way to make this work locally on you GPU, but it is quite a demanding effort I am afraid.

On Mon, 11 Mar 2019, 12:42 Rafik Gouiaa, notifications@github.com wrote:

@Johnqczhang https://github.com/Johnqczhang, I already installed this project. However, the frame rate is too low. A lot of persons had too low frame rate... So, I think it is fake project... just because it is coming from facebook

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/facebookresearch/DensePose/issues/4#issuecomment-471523386, or mute the thread https://github.com/notifications/unsubscribe-auth/Ackt3bCgjibFBcL4S6kxaBmHMDVHrAn4ks5vVk9EgaJpZM4UuFY2 .

rafikg commented 5 years ago

@Tetsujinfr Thanks for your answer. I forget about this project. Not worthy at all!

Tetsujinfr commented 5 years ago

Openpose is a lot easier to use and tweak in my view, although not delivering similar output. But pretty similar use cases or basic building blocks I think.

On Mon, 11 Mar 2019, 13:05 Rafik Gouiaa, notifications@github.com wrote:

@Tetsujinfr https://github.com/Tetsujinfr Thanks for your answer. I forget about this project. Not worthy at all!

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/facebookresearch/DensePose/issues/4#issuecomment-471530284, or mute the thread https://github.com/notifications/unsubscribe-auth/Ackt3TPfjq9-dmKO2BSMWQX1dCpvPSVOks5vVlSngaJpZM4UuFY2 .

SriRamGovardhanam commented 4 years ago

I managed to get it installed after quite a lot of troubles. I did re-install from a clean Ubuntu install. Did install Cuda 8.0 with Cudnn 7. I did compile caffe2 from source and used miniconda for the python and other dependency stack. My caffe2 installed libraries are under usr/local/lib which I did put in the LD_LIBRARY_PATH and my caffe2 modules are in the pytorch/build folder which I did add to my PYTHONPATH. I had to fix a small issue with ATen headers which were not installed properly (this will likely be fixed with newer versions of caffe2 code), I did mention the details under issue #97. Here is a test video I processed with the FB pre-trained model: https://youtu.be/apiwISDRB3k

Quite an effort to get this code to run but happy it works fine in the end.

hey , it would be so helpful to noobies like me , if you make a step by step tutorial about this densepose

jaggernaut007 commented 4 years ago

Use Docker

Best,

Shreyas Jagannath

+91 8792975919

http://www.jeytech.in

On Sat, Oct 19, 2019 at 1:06 AM sriram notifications@github.com wrote:

I managed to get it installed after quite a lot of troubles. I did re-install from a clean Ubuntu install. Did install Cuda 8.0 with Cudnn 7. I did compile caffe2 from source and used miniconda for the python and other dependency stack. My caffe2 installed libraries are under usr/local/lib which I did put in the LD_LIBRARY_PATH and my caffe2 modules are in the pytorch/build folder which I did add to my PYTHONPATH. I had to fix a small issue with ATen headers which were not installed properly (this will likely be fixed with newer versions of caffe2 code), I did mention the details under issue #97 https://github.com/facebookresearch/DensePose/issues/97. Here is a test video I processed with the FB pre-trained model: https://youtu.be/apiwISDRB3k

Quite an effort to get this code to run but happy it works fine in the end.

hey , it would be so helpful to noobies like me , if you make a step by step tutorial about this densepose

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/facebookresearch/DensePose/issues/4?email_source=notifications&email_token=AJPEPNF3UXOW7BCMSAQJQ6TQPIF2RA5CNFSM4FFYKY3KYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEBVVKMI#issuecomment-543905073, or unsubscribe https://github.com/notifications/unsubscribe-auth/AJPEPNFZCA26SLTTCWHPI53QPIF2RANCNFSM4FFYKY3A .