Training was done on google colab. The NB's are colab notebooks. To run the ipynb files, provide the required paths inside the notebook. The code is referenced from https://github.com/AnyiRao/SceneSeg the LGSS repo and architectural modifications are made on top of it
Results:
tt2488496.pkl
tt2582846.pkl
tt1707386.pkl
tt1412386.pkl
tt2024544.pkl
tt1375666.pkl
tt1205489.pkl
Scores: {
"AP": 0.519881324814132,
"mAP": 0.532751553052002,
"Miou": 0.528036302457737,
"Precision": 0.44037906827821616,
"Recall": 0.6041040503907537,
"F1": 0.5020978697516447
}
Pretrained Model: https://drive.google.com/file/d/102i2-kumZQlV42OrKXVVjvBPy66pHRfo/view?usp=sharing