Skip to content

Latest commit

 

History

History

Assignment-4

README.md

Assignment 4

Introduction

In this assignment, we worked on COCO dataset, which is a large-scale object detection, segmentation, and captioning dataset. We were required to to run 3 different models on the COCO dataset and compare the results.

Model 1: Faster R-CNN resnet50

  • Faster R-CNN is a state-of-the-art object detection model. It is based on the R-CNN model, which is a two-stage object detection model.
  • The first stage of R-CNN is a selective search algorithm, which generates a large number of region proposals.
  • The second stage is a classifier that classifies each region proposal into one of the 20 object classes.
  • The classifier is a linear SVM trained on the features extracted from the region proposals.
  • The R-CNN model is slow because it requires running the selective search algorithm on each image.
  • Faster R-CNN addresses this problem by using a deep convolutional network to generate region proposals.
  • The deep network is a fully convolutional network that takes an image as input and outputs a set of region proposals.
  • The region proposals are then classified by the same linear SVM classifier used in R-CNN.
  • The Faster R-CNN model is faster than R-CNN because it only requires running the deep network once per image.
  • The deep network is trained end-to-end, which means that it is trained jointly with the linear SVM classifier.

Model 2: FCOS (Fully Convolutional One-Stage Object Detection)

  • FCOS is a one-stage object detection model.
  • It is based on the FPN model, which is a feature pyramid network.
  • The FPN model is a deep convolutional network that takes an image as input and outputs a set of feature maps.
  • The feature maps are used to generate region proposals.
  • The region proposals are then classified by a linear SVM classifier.
  • The FPN model is trained end-to-end, which means that it is trained jointly with the linear SVM classifier.
  • FCOS is faster than Faster R-CNN because it is a one-stage model.
  • FCOS is also more accurate than Faster R-CNN because it uses a feature pyramid network to generate region proposals.

Model 3: SSD (Single Shot MultiBox Detector)

  • SSD is a one-stage object detection model.
  • It is based on the FPN model, which is a feature pyramid network.
  • The FPN model is a deep convolutional network that takes an image as input and outputs a set of feature maps.
  • The feature maps are used to generate region proposals.
  • The region proposals are then classified by a linear SVM classifier.
  • The FPN model is trained end-to-end, which means that it is trained jointly with the linear SVM classifier.
  • SSD is faster than Faster R-CNN because it is a one-stage model.
  • SSD is also more accurate than Faster R-CNN because it uses a feature pyramid network to generate region proposals.

Results

  • The results of the 3 models are shown in the following table:
Model mAP FPS
Faster R-CNN resnet50 0.xx x.x
FCOS 0.xxx x.x
SSD 0.xxx x.x

Conclusion

  • In this assignment, we worked on COCO dataset, which is a large-scale object detection, segmentation, and captioning dataset.
  • We ran 3 different models on the COCO dataset and compared the results.

References