Vehicle Attributes Model

#Work

• Vehicle Attribute Data Acquisition & Augmentation 🆔 npkahh ⏫

  • Fetch a vehicle detection dataset for road surveillance
  • Crop the images to fit the vehicle only.
  • Run the vlm on the dataset to generate attributes (vehicle color, vehicle type, vehicle brand).
  • Apply targeted data augmentation (e.g., color jittering, cropping, or mixing) specifically on underrepresented classes.

• [/] Train & Validate Vehicle Attributes Model 🆔 5el4gs ⛔ npkahh 🔼 Train the multi-task/attribute recognition model with the newly balanced dataset to boost recall on the target attributes.

  • Modify notebook parameters for vehicle model.
  • Run training on vehicle dataset.
    • Improve model architecture to fix val loss increase (overfitting)
    • Test other model architectures for this use case
    • Disable heads with insufficient classifications
    • Add class imbalance weighting to counter bias
    • Add a lr tuning connected to metrics and not max epochs
    • [/] Build another dataset with a higher frameskip mobilenetv3-small came out better than the large model, with resnet18 comin in second place compared resnet18, mobilenetv3-small, mobilenetv3-large, effecientnet.
  • Export model to onnx
  • Add the model to infer sever (Take person attribute model as reference)
  • Add a custom parser cpp file for vehicles
  • Integrate into deepstream pipeline