berkeley_deepdrive_experimentation

Requirements

Dataset Preparation from Berkeley Deep Drive to shortlist 1000 good quality images (900 for train and 100 for validation).
Fine-tuned a FasterRCNN model with specific target labels
Evaluate by issuing sample queries to a test video taken from the same dataset

Sample query: "Sample every 3 seconds and get all timestamps where there are more than 2 cars, 1 sign, 1 pedestrian"
Momentary change in border color to green indicates the frame has satisfied the query

Name		Name	Last commit message	Last commit date
Latest commit History 6 Commits
models		models
README.md		README.md
coco_eval.py		coco_eval.py
coco_utils.py		coco_utils.py
dataset.py		dataset.py
dataset_preparation.ipynb		dataset_preparation.ipynb
engine.py		engine.py
model_test_1.gif		model_test_1.gif
train.ipynb		train.ipynb
transforms.py		transforms.py
utils.py		utils.py
video_queries.ipynb		video_queries.ipynb