car-dataset / README.md
shravya11's picture
Upload README.md with huggingface_hub
b00b628 verified
---
license: other
task_categories:
- image-classification
- object-detection
tags:
- computer-vision
- vehicles
- car
- cars
- raw-images
---
# Car Image Dataset
This repository contains an expanded dataset of **10,765 car images** divided into standard splits (training, validation, testing), additional training data, and a RAW sensor/ISP sub-dataset with object detection annotations.
- **Total Images**: 10,765
- **Dataset Structure**:
- `train/`: 4,272 images total:
- 920 flat, unstructured car images (directly in the root)
- 3,352 structured images divided into 7 brand class folders:
- `Audi/`: 814 images
- `Hyundai Creta/`: 271 images
- `Mahindra Scorpio/`: 316 images
- `Rolls Royce/`: 311 images
- `Swift/`: 424 images
- `Tata Safari/`: 441 images
- `Toyota Innova/`: 775 images
- `train_additional/`: 5,000 flat, unstructured car images (extracted from archive 4, split into two chunks to avoid API timeouts):
- `part_1/`: 2,500 images
- `part_2/`: 2,500 images
- `validation/`: 230 flat, unstructured car images
- `test/`: 813 structured images divided into 7 brand class folders:
- `Audi/`: 199 images
- `Hyundai Creta/`: 67 images
- `Mahindra Scorpio/`: 75 images
- `Rolls Royce/`: 74 images
- `Swift/`: 102 images
- `Tata Safari/`: 106 images
- `Toyota Innova/`: 190 images
- `raw_dataset/`: 450 images total representing 225 unique car scenes under raw processing conditions:
- `isp_processed/`: 225 JPEG images processed using standard Image Signal Processor pipelines.
- `raw_converted/`: 225 lossless PNG images converted from the raw sensor format (scaled [0, 255] and edge-pixels removed).
- `annotations.xls`: Bounding box annotations for the 225 scenes in the format `[top_left_x, top_left_y, box_height, box_width]`.
## Brand Classes
- Audi
- Hyundai Creta
- Mahindra Scorpio
- Rolls Royce
- Swift
- Tata Safari
- Toyota Innova
## License
License details for this dataset are unspecified.