Training code release about romp HOT 19 CLOSED

Thanks for your interests!
The paper is still under submission.
The training code will be released when the paper is accepted.

from romp.

Really appreciate your attension! Sorry, I don't know the exact data. But I believe it won't be too long.
I will try my best to get the code released as soon as possible.

from romp.

Great thank you.
I really like that your approach is bottom-up and specially the idea of repulsion fields of the body centers!
Hope it is released soon. Do you know the date of the acceptance announcement?

from romp.

We are really sorry for the delayed release of training code. ROMP just gets in ICCV 2021. I am preparing the training code (on fire). There are multiple version of ROMP training code. I will release the code for fine-tunning first in convenience of a fast try. My appology.

from romp.

We are really sorry for the delayed release of training code. ROMP just gets in ICCV 2021. I am preparing the training code (on fire). There are multiple version of ROMP training code. I will release the code for fine-tunning first in convenience of a fast try. My appology.

Awesome!

from romp.

The training code has been released, along with almost all evaluation protocols~

from romp.

Hi @Arthur151 ,

Congratulations for this awesome work and thank you so much for sharing this code!
Concerning training, I noticed a required file "datasets/h36m/h36m_train.txt" that I was not able to find yet.
I checked the source code, ZIP files in the releases and Google Drive, without success. So I suppose this file should be in the original H36M dataset, right?
Otherwise, can it be generated just considering all frames from the training subjects?

Thanks.

from romp.

As an additional comment: I just noticed that the file "datasets/mpii/annot/train.json" is also required for training. Maybe I missed something in the installation process?

from romp.

Sorry, my mistake. I have uploaded these files to the google drive.
Please be aware to change this line.

Don't hestitate to contact me if there are any issues. I will fix them as soon as possible.

from romp.

@Arthur151 "datasets/h36m/h36m_train.txt" still not exist by now, while the google drive is 404 error. And when I "python -m romp.lib.dataset.mpi_inf_3dhp", I find "dataset/mpi_inf_3dhp/S1/Seq1/annot.mat " is still needed, while your google drive share the different directory structure of mpi-inf-3dhp dataset. Please check it out.

from romp.

We have change the link to google drive.

from romp.

Please be aware that this line will be processed if only you miss the annots.npz file. This function is just used to prepare the annots.npz.
You can download it from our google drive.

from romp.

@liangwx Anything else? Come on. 我能打十个。

from romp.

Yes hh, When I add the "datasets/h36m/h36m_train.txt", I run "python -m romp.lib.dataset.h36m" again, but get error like this:
Exception has occurred: TypeError
Caught TypeError in DataLoader worker process 0.
Original Traceback (most recent call last):
File "/home/zhaohan/Install/anaconda3/envs/romp-py37-cu111/lib/python3.7/site-packages/torch/utils/data/_utils/worker.py", line 287, in _worker_loop
data = fetcher.fetch(index)
File "/home/zhaohan/Install/anaconda3/envs/romp-py37-cu111/lib/python3.7/site-packages/torch/utils/data/_utils/fetch.py", line 44, in fetch
data = [self.dataset[idx] for idx in possibly_batched_index]
File "/home/zhaohan/Install/anaconda3/envs/romp-py37-cu111/lib/python3.7/site-packages/torch/utils/data/_utils/fetch.py", line 44, in
data = [self.dataset[idx] for idx in possibly_batched_index]
File "/home/zhaohan/wxl/ROMP/romp/lib/dataset/image_base.py", line 342, in getitem
return self.get_item_single_frame(index)
File "/home/zhaohan/wxl/ROMP/romp/lib/dataset/image_base.py", line 85, in get_item_single_frame
info = self.get_image_info(index)
File "/home/zhaohan/wxl/ROMP/romp/lib/dataset/h36m.py", line 86, in get_image_info
image = cv2.imread(imgpath)[:,:,::-1]#把BGR转成RGB
TypeError: 'NoneType' object is not subscriptable
File "/home/zhaohan/wxl/ROMP/romp/lib/dataset/image_base.py", line 492, in test_dataset
for _,r in enumerate(dataloader):
File "/home/zhaohan/wxl/ROMP/romp/lib/dataset/h36m.py", line 123, in
test_dataset(h36m, with_3d=True,with_smpl=True,)

from romp.

This error is pretty obvious. You can try to print the imgpath in image = cv2.imread(imgpath)[:,:,::-1].
Please make sure the imgpath refer to the correct path.
I guess that you didn't put the h36m dataset at the right place. Or the name of the image folder is wrong.
To correct this, please check this line and this line

from romp.

@liangwx Come on. I am on fire.

from romp.

@Arthur151 Sorry to reply so late, maybe you'd like to use Relative Path other than Absolute one in h36m_train.txt. And mayby naming the image files without "spacing" could be more compatible with my ubuntu envirenment.

from romp.

this line output this :
'/export/home/suny/dataset/h36m/images/S6_Smoking 1_3_49.jpg'

from romp.

I see. Thanks for the bug report. I have update the h36m_train.txt in google drive.
欢迎再次光临。

from romp.