Novelty Detection using CNN on UCSD-PED dataset
Demo image
Link to Code:
Its around 11GB project, so I have not uploaded the code anywhere.
How I built it
It was a cool experience to work on an unsupervised machine learning. But, sometimes I felt it was very hectic. I read many papers about the unsupervised models, novely detection, anomaly detection. Also found some same implementations but on other benchmark datasets like MNIST and CIFAR-10. So, I opted the UCSD-ped dataset for my project. I needed a GPU for my cuda functions to work. But without it, I started the project, used my 12 gig ram to train without the cuda. I thought I was dumb choosing this project. But later, my counselor helped me and provided me the access to the university lab resources. And this way, I completed my project on a roller-coaster ride.
Additional Thoughts / Feelings / Stories
Whenever you feel like dropping any idea of project, talk to someone of the same field. Maybe he/she cannot give you better ideas to optimize/upgrade your project but you might end up with path leading to solution of your problem.
Top comments (0)