metadata

license: mit
library_name: pytorch
tags:
  - faster-rcnn
  - object-detection
  - computer-vision
  - pytorch
  - bdd100k
  - autonomous-driving
  - BDD 100K
  - fine-tuned
  - hallucination-mitigation
  - out-of-distribution
pipeline_tag: object-detection
datasets:
  - bdd100k
widget:
  - src: >-
      https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/bounding-boxes-sample.png
    example_title: Sample Image
model-index:
  - name: faster-rcnn-bdd-finetune
    results:
      - task:
          type: object-detection
        dataset:
          type: bdd100k
          name: Berkeley DeepDrive (BDD) 100K
        metrics:
          - type: mean_average_precision
            name: mAP
            value: TBD

Faster R-CNN - Berkeley DeepDrive (BDD) 100K Fine-tuned

Faster R-CNN model fine-tuned on Berkeley DeepDrive (BDD) 100K dataset to mitigate hallucination on out-of-distribution data in autonomous driving scenarios.

Model Details

Model Type: Faster R-CNN Object Detection
Dataset: Berkeley DeepDrive (BDD) 100K
Training Method: fine-tuned to mitigate hallucination on out-of-distribution data
Framework: PyTorch
Task: Object Detection

Dataset Information

This model was trained on the Berkeley DeepDrive (BDD) 100K dataset, which contains the following object classes:

car, truck, bus, motorcycle, bicycle, person, traffic light, traffic sign, train, rider

Dataset-specific Details:

Berkeley DeepDrive (BDD) 100K Dataset:

100,000+ driving images with diverse weather and lighting conditions
Designed for autonomous driving applications
Contains urban driving scenarios from multiple cities
Annotations include bounding boxes for vehicles, pedestrians, and traffic elements

Usage

This model can be used with PyTorch and common object detection frameworks:

import torch
import torchvision.transforms as transforms
from PIL import Image

# Load the model (example using torchvision)
model = torch.load('path/to/model.pth')
model.eval()

# Prepare your image
transform = transforms.Compose([
    transforms.ToTensor(),
])

image = Image.open('path/to/image.jpg')
image_tensor = transform(image).unsqueeze(0)

# Run inference
with torch.no_grad():
    predictions = model(image_tensor)

# Process results
boxes = predictions[0]['boxes']
scores = predictions[0]['scores']
labels = predictions[0]['labels']

Model Performance

This model was fine-tuned to mitigate hallucination on out-of-distribution data on the Berkeley DeepDrive (BDD) 100K dataset using Faster R-CNN architecture.

Fine-tuning Objective: This model was specifically fine-tuned to mitigate hallucination on out-of-distribution (OOD) data, improving robustness when encountering images that differ from the training distribution.

Architecture

Faster R-CNN (Region-based Convolutional Neural Network) is a two-stage object detection framework:

Region Proposal Network (RPN): Generates object proposals
Fast R-CNN detector: Classifies proposals and refines bounding box coordinates

Key advantages:

High accuracy object detection
Precise localization
Good performance on small objects
Well-established architecture with extensive research backing

Intended Use

Primary Use: Object detection in autonomous driving scenarios
Suitable for: Research, development, and deployment of object detection systems
Limitations: Performance may vary on images significantly different from the training distribution

Citation

If you use this model, please cite:

@article{ren2015faster,
  title={Faster r-cnn: Towards real-time object detection with region proposal networks},
  author={Ren, Shaoqing and He, Kaiming and Girshick, Ross and Sun, Jian},
  journal={Advances in neural information processing systems},
  volume={28},
  year={2015}
}

License

This model is released under the MIT License.

Keywords

Faster R-CNN, Object Detection, Computer Vision, BDD 100K, Autonomous Driving, Deep Learning, Two-Stage Detection