search

Mandroide / gpasb-nvidia2021-track4

Customized Implementation of "Good Practices and A Strong Baseline for Traffic Anomaly Detection"
MIT License
3 stars 3 forks source link

Running your fork of kevaldoshi17/NVIDIA_AICITY produced empty result #1

Closed hieplpvip closed 2 years ago

hieplpvip commented 3 years ago

Hi,

I'm having some problems running your fork of kevaldoshi17/NVIDIA_AICITY. I know this is not the right repo, but I can't create issues for a fork, and this repo is of the same topic, so I think it's appropriate to create an issue here.

I'm running your code with the 2019 track 3 testset (not sure if it's a problem, but as far as I know the datasets don't change between years). I only made some minor changes to extract_frames.py to not load DOT Iowa Accident Labels. You can find my working notebook here: https://github.com/hieplpvip/traffic-anomaly-detection-baseline/blob/master/notebook.ipynb. I got an empty result.csv. The *.npy files are also very small:

4.0K    bounds1.npy
4.0K    bounds2.npy
12K     centers1.npy
4.0K    centers2.npy
80K     change.npy
4.0K    result1.npy
4.0K    result2.npy

I don't really know what to do now (been stuck for a few days). Could you take a look at my notebook? Thanks a lot!

Mandroide commented 3 years ago

Hi,

I'm having some problems running your fork of kevaldoshi17/NVIDIA_AICITY. I know this is not the right repo, but I can't create issues for a fork, and this repo is of the same topic, so I think it's appropriate to create an issue here.

I'm running your code with the 2019 track 3 testset (not sure if it's a problem, but as far as I know the datasets don't change between years). I only made some minor changes to extract_frames.py to not load DOT Iowa Accident Labels. You can find my working notebook here: https://github.com/hieplpvip/traffic-anomaly-detection-baseline/blob/master/notebook.ipynb. I got an empty result.csv. The *.npy files are also very small:

4.0K    bounds1.npy
4.0K    bounds2.npy
12K     centers1.npy
4.0K    centers2.npy
80K     change.npy
4.0K    result1.npy
4.0K    result2.npy

I don't really know what to do now (been stuck for a few days). Could you take a look at my notebook? Thanks a lot!

To avoid using the version with spreadsheet, just checkout the commit before "Add ground truth anomaly".

I have tried out the code before, and is not able to detect anomalies. I don't know if there is a modification that went wrong in the fork or the author's algorithm does not work correctly. The Result.csv was just a copy & paste from Result.txt the author did.

hieplpvip commented 3 years ago

Thanks! I'll try running the original repo to see if it's working.

Btw, I see you pinned your fork of selab-hcmus/AI_City_2020. Does that mean you have managed to get it working?

Mandroide commented 3 years ago

Thanks! I'll try running the original repo to see if it's working.

Btw, I see you pinned your fork of selab-hcmus/AI_City_2020. Does that mean you have managed to get it working?

I haven't gotten the first functional anomaly detector yet. Most of them are hardcoded, missing stuff or just don't work.

I have mentioned something like that in Suggestions for upcoming code.