I’m gonna make it simple for you guys

Detectron 2😍YOLOv5😔
BackboneFaster R-CNNOwn Architecture
DatasetBooks – 83 ImagesBooks – 83 Images
Dataset formatCOCOYOLO
Training Time1Hr 15Mins>2Hrs
Accuracy99% – Impressive>90% accuracy achieved only for few images, other images with <90% accuracy but not too low
InferenceThe Best model with higher detection rate even for Low-resolution images and noisy ones too – Author’s favoriteOutput look is good (Bounding Box feature), Higher accuracy for single class images in few minutes
NoteThe dataset must be in a high resolutionBlurred image has been detected
Table for Comparison between our models

I have missed few points, let’s discuss and review them

Why we use COCO format dataset?

The answer is COCO format comes with only 80 similar objects, if you have the dataset from COCO – It’s well and good, You can train it directly but if you’re trying the model with your dataset, you have to register the dataset and it’s labels

R-CNN family🤔

The goal is to process the input images and produce a set with bounding box output. Each bounding box contains an object and the category (car or pedestrian). More recently, R-CNN has been extending to perform other computer vision tasks. The following covers some of the versions of R-CNN that are developing

  • R-CNN
  • FASTER R-CNN (Detectron 2)
  • MASK R-CNN for instance segmentation and object detection

The R-CNN family is suitable for Multi-Class Object detection


I might learn VIDEO OBJECT DETECTION, IMAGE SEGMENTATION on images, CNN and post as future articles

Leave a Reply

Please log in using one of these methods to post your comment:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s