artest08 / latetemporalmodeling3dcnn Goto Github PK

Hi,
Thank you for your great work!
I'm currently trying to run the training for the the baseline code using "two_strream_bert2.py"
I get the following error:
ModuleNotFoundError: No module named 'weights.model_path'

The weights folder contains 5 files which I downloaded as per the instructions. However, the "model_path.py" file is apparently missing from this folder. Therefore, the following import fails:
"from weights.model_path import rgb_3d_model_path_selection".

Please could you point me to where I could get this file? or is there an alternative to fixing this issue?

Thanks

Hi. This is a great job. I'm just a beginner to this work. I just wonder if a video contains 2 or more actions so the model can automatically separate them and give us a correct prediction or we must do preprocessing ? If preprocess, any suggestion for that ?
Thank you so much !!!

Hi! I am training your awesome model in ucf101. Could you share the training parameters' detail for R(2+1)D BERT (32f) R(2+1)D BERT (64f).

I read README.md and then found it mentioned that 'Later, please download the necessary files from the link, and copy them into the main directory. https://1drv.ms/u/s!AqKP51Rjkz1Gaifd54VbdRBn6qM?e=7OxYLa',however I just cannot open this link.Could you please offer a new link?Thank you very much!

Thank you for this great work.

I am working on a similar problem, I want to apply BERT for some frame features extracted using I3D, but I want to perform frame-level classification rather than video classification. I was wondering how I can adapt your implementation since you use the classification token which is defined on video-level and not frame-level.

I also had a question regarding the training of the model, I wanted to know why you don't include the loss from the predictions of the masked frames?

Any help would be appreciated 😄 !

Hi !! I am working on Accident Detection from CCTV Cameras. It will be helpful for my research and will also save lot of computation time if you can provide me the pretrained model of R2+1D-BERT. Thank you!!

Hello. First of all, congratulations on your work.

I want to reproduce your experiments, but I am stuck on the data preprocessing phase. The original dataset comes as ".avi" videos, how do you extract the RGB and optical flow? How much frames? And how do you convert from the original file-per-class split files to the structure you have in the "datasets/settings/hmdb51/" folder?

Disclaimer: I am just now getting started with action recognition research, this might be standard procedure for any other implementation using the HMDB51 dataset, but I am not yet familiarized with it.

I appreciate any help you can provide.

Hi Great work you've done here!
Please how do I test the model on long video clip with different actions occurring at separate time intervals?
Thanks

• add arxiv url of the paper
• add bibtex citation
• add model url table (optional)

Hi, I am reading your paper and code in the past few days. I found the code and the paper are inconsistent. One big part of the paper is the removal of the Temporal Global Average Pooling in Figure 1. For example, in your rgb_I3D.py code, in the model rgb_I3D64f_bert2, the input dimension to 3DCNN is batch x 3 x 64 x 224 x 224, the output is batch x 1024 x 8 x 7 x 7. Then you apply another 3D pooling to get batch x 1024 x 8 x 1 x 1. In my understanding, the temporal pooling is already done in the 3DCNN. In Figure 1 of your paper, you remove the Temporal Global Average Pooling and the output of the 3DCNN still has f1,f2,...,fN. But in your code, there is no such N frame features. Can you help me understand your code and paper?
Thanks a lot.

add python scripts to automatically download model weights.

In two_stream_bert2.py, I believe there is a typo in an import statement (line 33):
from weights.model_path import rgb_3d_model_path_selection
should be
from utils.model_path import rgb_3d_model_path_selection

I get the following error when I try to run the code: I tried to fix that, but can't understand the problem.

Traceback (most recent call last): File "two_stream_bert2.py", line 33, in <module> from weights.model_path import rgb_3d_model_path_selection ModuleNotFoundError: No module named 'weights.model_path'

when I try to create the environment this error shows up
ERROR: No matching distribution found for fvcore==0.1 (from -r /home/afrina/LateTemporalModeling3DCNN-master/condaenv.63nxe8d0.requirements.txt (line 13))

Thank for your awesome repo.
Could you please share the data frames through my link:

https://drive.google.com/drive/folders/1EKtOvnXW3PLPfJsCBaSB5AWIJIldlFcL?usp=sharing

thank you so much.

I tried your eval script (spatial_demo_bert.py) it worked at first but after a while a Cuda runtime error occurred. The first two or three times the error was gone after creating a new environment but now it produces the error no matter what I do.

I have a NVIDIA GeForce RTX 2070 SUPER. I am working on Windows.

Thank you in advance

While running conda env create -f LateTemporalModeling3D.yml getting following error

ResolvePackageNotFound:
  - fontconfig==2.13.0=h9420a91_0
  - pcre==8.43=he6710b0_0
  - scikit-learn==0.20.1=py36h4989274_0
  - mkl_fft==1.0.4=py36h4414c95_1
  - libgcc-ng==9.1.0=hdf63c60_0
  - libstdcxx-ng==8.2.0=hdf63c60_1
  - wrapt==1.11.2=py36h7b6447c_0
  - tornado==6.0.3=py36h7b6447c_0
  - icu==58.2=h9c2bf20_1
  - psutil==5.6.7=py36h7b6447c_0
  - zeromq==4.3.1=he6710b0_3
  - wurlitzer==2.0.0=py36_0
  - ncurses==6.1=he6710b0_1
  - lazy-object-proxy==1.4.3=py36h7b6447c_0
  - gmp==6.1.2=h6c8ec71_1
  - pyqt==5.9.2=py36h05f1152_2
  - glib==2.63.1=h5a9c865_0
  - dbus==1.13.12=h746ee38_0
  - tk==8.6.8=hbc83047_0
  - xz==5.2.4=h14c3975_4
  - pyrsistent==0.15.6=py36h7b6447c_0
  - libuuid==1.0.3=h1bed415_2
  - libtiff==4.0.9=he85c1e1_1
  - libsodium==1.0.16=h1bed415_0
  - typed-ast==1.4.0=py36h7b6447c_0
  - ptyprocess==0.6.0=py36_0
  - freetype==2.9.1=h8a8886c_1
  - mkl_random==1.0.1=py36h4414c95_1
  - qt==5.9.6=h8703b6f_2
  - libffi==3.2.1=hd88cf55_4
  - zlib==1.2.11=h7b6447c_3
  - libedit==3.1.20181209=hc058e9b_0
  - libgfortran-ng==7.2.0=hdf63c60_3
  - libpng==1.6.37=hbc83047_0
  - expat==2.2.6=he6710b0_0
  - readline==7.0=h7b6447c_5
  - ujson==1.35=py36h14c3975_0
  - gstreamer==1.14.0=hb453b48_1
  - pyzmq==18.1.0=py36he6710b0_0
  - python==3.6.5=hc3d631a_2
  - mistune==0.8.4=py36h7b6447c_0
  - openssl==1.0.2t=h7b6447c_1
  - cryptography==2.3.1=py36hc365091_0
  - cffi==1.13.2=py36h2e261b9_0
  - jpeg==9b=h024ee3a_2
  - markupsafe==1.1.1=py36h7b6447c_0
  - libxml2==2.9.9=hea5a465_1
  - secretstorage==3.1.1=py36_0
  - sqlite==3.30.1=h7b6447c_0
  - libspatialindex==1.9.3=he6710b0_0
  - sip==4.19.8=py36hf484d3e_0
  - yaml==0.1.7=had09818_2
  - pywavelets==1.0.2=py36hdd07704_0
  - libxcb==1.13=h1bed415_1
  - gst-plugins-base==1.14.0=hbbd80ab_1

I have seen this code:
input_vectors=x
norm = input_vectors.norm(p=2, dim = -1, keepdim=True)
input_vectors = input_vectors.div(norm)
Is that code's function same as Batch Normalization?

Thank you so much for sharing the idea.

An error occurred when I execute the following command：conda env create -f LateTemporalModeling3D.yml

ERROR: Could not find a version that satisfies the requirement roi-align==0.0.2 (from -r /data/LateTemporalModeling3DCNN/condaenv.u7mj_9s2.requirements.txt (line 58)) (from versions: none)
ERROR: No matching distribution found for roi-align==0.0.2 (from -r /data/LateTemporalModeling3DCNN/condaenv.u7mj_9s2.requirements.txt (line 58))

CondaValueError: pip returned an error

Thanks for your help!