Hi, is it possible to release a requirements.txt with version for each library? I am currently trying to build Detectron and I have been running into this error:

    with_cuda=with_cuda)
  File "/home/siteng/anaconda3/envs/test/lib/python3.6/site-packages/torch/utils/cpp_extension.py", line 1255, in _write_ninja_file_and_compile_objects
    error_prefix='Error compiling objects for extension')
  File "/home/siteng/anaconda3/envs/test/lib/python3.6/site-packages/torch/utils/cpp_extension.py", line 1555, in _run_ninja_build
    raise RuntimeError(message) from e
RuntimeError: Error compiling objects for extension

I am speculating that there may be some kind of compatibility issue. Thank you!

I find this an interesting work and I would like to reimplement your work following the code. However, I find that I cannot access the link you provided in readme for features for VIDVRD dataset(https://internal-api-drive-stream.larksuite.com/space/api/box/stream/download/all/boxusS8Z0kwEizoPPh5h7vx7Usf/). Can I know is there any other way for me to access it? Thanks in advance

Thank you for your awesome work and I enjoy reading it. When I tried to run your code, I found two questions:

1. How long does it spend to train on AG dataset under the PredCLS settings? I run test_net_rel.py with one 3090 GPU finding that the run_inference process would take about 6 hours and that's just the test process. May I ask how long the whole training process took?
2. For each center frame in the same video, the model would compute the temporal features with I3D. However, the sampled frame segment could be highly similar for different center frames from the same video. I wonder that if some of the I3D computation is redundant?

Hello, the author, thank you very much for your excellent work of open source. When I used the pre training model to test, I encountered the following problems. How can I solve them? I want to evaluate for detected boxes, Recalls (SGDet) dataset AG

Hi, I tried running the evaluation script of the VidVRD pretrained model, but the documentation is a little vague in this regard. At the moment I'm using the dataset mentioned but I want to evaluate custom videos too. Still, using the mentioned dataset, I always get the error of missing baseline_relation_prediction.json file in the Output folder. I am using Google Colab to run the code and hence changed the directories paths in the code scripts. Can you tell me why isn't the baseline relation JSON file not being created in the output folder, what could be the reason? Because I tried the paths in every script we're calling. I think there is some problem with the CUDA availability because it didn't run with the CUDA_VISIBLE_DEVICES=1,2,3,4,5,6,7 (in the evaluation script) so I tried changing it to CUDA_VISIBLE_DEVICES=1 and it still didn't work so I used CUDA_VISIBLE_DEVICES=0. Can you please take a look

Thank You

AssertionError: Annotation file '/vision2/u/zhuoyih/projects/moma_scene_graph/TRACE/data/ag/annotations/detections_train.json' not found

Hi, I got this error while I was trying to train on the action genome dataset, however, I searched there's no detections_train.json in action genome dataset or ways to generate this file, could you give more information on this, please?

Hi, dear author, i want to obtain the video scene graphs on custom videos (e.g. the videos from ActivityNet dataset), could you provide some script file to guide my research?

Thanks for sharing the nice work!

But I find the performance presented in the paper is very different with methods in "Action Genome: Actions as Compositions of Spatio-temporal Scene Graphs" and "detecting human-object relationships in videos". What are the causes for this?