Topic: Weakly supervised object detection using CNN Course project for CS676A:Computer Vision at IIT Kanpur. Please find the report, poster and the presentation in the same repository. Papers followed: 1) Oquab, Maxime, et al. "Is object localization for free?- weakly-supervised learning with convolutional neural networks." Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2015. 2) Oquab, Maxime, et al. "Learning and transferring mid-level image representations using convolutional neural networks." Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2014. 3) Lempitsky, Victor, and Andrew Zisserman. "Learning to count objects in images." Advances in Neural Information Processing Systems. 2010. Group members: L S Vishnu Sai Rao (12376) and Saurabh Kataria (12637) Platform and tools used: Torch7 - https://github.com/torch/torch7 Linux MATLAB Nvidia GPU and its CUDA Datasets used: PASCAL VOC 2012 PASCAL VOC 2007 Dependencies of code: Common libraries of Torch7 OpenBLAS Nvidia CUDA latest 6.5+ Directions to execute the code: 1) Run 'generatedataset-trainval-full-weak-500.lua' and 'generatedataset-test-full-weak-500.lua' to generate training and testing data in t7 format. 2) Run 'post-train.lua' on a system with capable GPU and memory to train the CNN model. 3) Run 'run-on-test-set.lua' to create maxpooledscores (output) by testing the trained model on test data (in t7 format). 4) Run 'confusionmatrix.lua' to generate the accuracies of all classes calculated from the confusion matrices obtained for each class. Index of brief description of each script: 1) confusionmatrix.lua - generates the confusion matrix for each class and calculate the accuracies too. 2) featureextractor.lua - extracts the features from an image and store it into t7 format 3) generatedataset-test-full-weak-500.lua - converts test image set into batches of t7 format 4) generatedataset-trainval-full-weak-500.lua - converts training image set into batches of t7 format 5) post-train.lua - trains the CNN 6) run-on-test-set.lua - create maxpooledscores (output) by testing the trained model on test data (in t7 format)