The problem To detect dents/scratches from images of cars solves

To segment/detect/recognize dents/scratches from images of cars. The dataset contains car images with one or
more damaged parts. The img/ folder has all 80 images in the dataset. There are two more folders train/ and Val/
for training, validation and testing purposes respectively.
Folders:
train/:
Contains 59 images.
COCO_train_annos.json: Train annotation file for damages where damage is the one and only category.
COCO_mul_train_annos.json: Train annotation file for parts having damages. There are five categories of parts
based on which part the damage has happened.
The parts can be namely, headlamp, front bumper, hood, door, rear bumper.
val/:
Contains 11 images.
COCO_val_annos.json: Validation annotation file for damages where damage is the one and only category.
COCO_mul_val_annos.json: Validation annotation file for parts having damages. There are five categories of
parts based on which part the damage has happened.
The parts can be namely, headlamp, front bumper, hood, door, rear bumper

HAD CREATED A SOFTWARE PROGRAM TO DETECT/SEGMENT THE
SCRATCHES AND DENTS IN A CAR USING THE DETECTRON(A META
AI WHICH IS DEVELOPING BY FACEBOOK)
CAN BE USED IN AUTOMOBILE INDUSTRY OR CAR
MANUFACTURING INDUSTRIES TO TEST OR TO VIEW AND ANALYSE
A CAR WITH DENTS SCRATCHES

Challenges we ran into

Detectron2 is a ground-up rewrite of Detectron
that started with maskrcnn-benchmark. The
platform is now implemented in PyTorch. With a
new, more modular design, Detectron2 is flexible
and extensible, and able to provide fast training
on single or multiple GPU servers. Detectron2
includes high-quality implementations of state-ofthe-art object detection algorithmsFirst of this we
had tried using the TensorFlow but Due to the
unavailability of necessary resources and error
occurring in the TensorFlow we choose the
detectron Where it was so helpful and the
precision and the Accuracy rate were a boost
compared to TensorFlow and it Was the
emerging technology in the developing world