Object detection using SSL Classification

Preamble 📜

The Proceedings of the European Conference on Computer Vision has conducted a benchmark PASCAL Visual Object Challenge (VOC) evaluating performance on object class recognition (from 2005-2012, now finished). For our task, we examine the 2007 dataset which consists of several types of random images collected in January 2007 from Flickr. There are five challenges: classification, detection, segmentation, action classification, and person layout.

Our goal from this challenge is to perform image classification from several visual object classes in realistic scenes (i.e. not pre-segmented objects). And so, we will be using certain Semi-Supervised Learning approaches where we have both labeled and unlabeled sets to check if we get superior results as opposed to supervised techniques.

Mandatory features

• MPEG-7 Color Layout Descriptor
• Visual Bag-of-Words
• Speeded up robust features (SURF)

Supplementary features

• Local Binary Patterns
• Color Histogram

Folder Structure 📂

Folder structure and naming conventions for this project

A top-level directory layout

.
├── HLD_LLD                  # Documentation files (.pdf)
├── codebase                  # workable codebase (.py, .ipynb)
├── fig
      └── plots                 # Compiled image files (.png, .jpg)
├── files                       # Legacy files                  
├── posterbase                  # layout and template files (.sty, .tex)                                      
├── .gitignore
├── LICENSE
├── README.md
└── requirements.txt          # Python libraries for this project (versioned)

Usage

Code requires Python 3.6 to run. Install the dependencies required for the code to run.

pip install -r requirements.txt

For running the code

python {change_the_model_here}.py --data ..\VOCtrainval_06-Nov-2007\VOCdevkit\VOC2007\

Our Findings in form of a Scientific Poster

Download in PDF

Datasets and References🌐

• F. Perronnin, J. Sánchez and Yan Liu, "Large-scale image categorization with explicit data embedding," 2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2010, pp. 2297-2304, doi: 10.1109/CVPR.2010.5539914.

• Everingham, L. Van Gool, C. K. I. Williams, J. Winn, and A. Zisserman. The PASCAL Visual Object Classes Challenge 2007 (VOC2007) http://host.robots.ox.ac.uk/pascal/VOC/voc2007/index.html

Contributors 👨‍💻

• Shubham Kumar Agrawal
• Pavan Tummala
• Usama Ashfaq
• Syed Muhammad Laique Abbas
• Ranji Raj