Blocking or removal of objectionable content involving minors in an attempt to reduce child abuse
Mashrin Srivastava (~Mashrin_Srivastava) |
The talk will be on the classification and removal of the offensive content which involves children, with a plan to extend it in a way that an extension can be developed. Websites and browsers can implement this to filter out any such content at the time of upload. Removal of such content, will lead to a scenario of decreased child abuse and sexual harassment as the content uploaded by the assaulter will be blocked at that instance and hence, such shameful acts might get reduced as in this case, one will not be able to access their content, thus nullifying their motto.
Identifying and classifying offensive or adult content is an important problem which researchers have tackled for decades. With the evolution of computer vision and deep learning, algorithms have matured and are now able to classify an image with greater precision. Since images and user generated content dominate the internet today, filtering nudity and other objectionable adult contents become an important problem. Using a deep CNN with residual connections, the model will quickly classify each second of a video into different categories, further also predict an expected age for the said content. Then it uses the same classification to automatically edit the video as per the government rules in that geographical area. It can remove all the scenes containing objectionable content. The proposed model on which the talk will be focused will be using a Convolution neural network with residual connections pre-trained on on ImageNet. The model will be iteratively fine tuned to enhance the accuracy. For the removal of the offensive content by cutting it out can be done in a way that with the predictions for a frame each second, it will take the argmax of those predictions and will create cut blocks of the content or video where the offensive content score is greater than some threshold. The threshold will be flexible, so that along with the classification, we can determine whether the content needs to be removed or not, according to the geographical location. The CNN will also classify the people on the basis gender and on the basis of age- as minor or adults.
- The basic idea about what Deep Learning is.
- Current state of affairs regarding online interaction and abuse.
- An urge to bring in a change in the "digital society" and the motivation to contribute towards what's right!
Mashrin is passionate about data science, algorithms and graphs and is presently in the final year pursuing Computer Science and Engineering from Vellore Institute of Technology. I have interned as a data scientist at Wingify, Cerelabs, Sales and Quotes and Knolskape and have been a contributor to the Stanford Scholar Initiative and processing.py
Saumya is a data science enthusiast currently working as Google Summer of Code student @GreenNavigation. She is presently a final year student pursuing Computer Science and Engineering from Vellore Institute of Technology and have previously interned with Cisco, Knolskape and Cerelabs along with contributing to Stanford Scholar initiative.
Sanjay is presently a final year student of Computer Science and Engineering at VIT University. He is extremely proficient in algorithms and web programming. He has a new found interest in Machine Learning and is planning to pursue it for higher education.
P.S.- Special thanks to Andrew Ng from all three of us (and many others from the student community)
GSoC link: https://github.com/Greennav/machine-learning