Abstract | ||
---|---|---|
Recently, supervised learning has witnessed efficient employment in computer vision problems. But, its utility in various domain is limited by its reliance on massive amount of labeled data. This conflicts with human visual learning which requires just a few views of the object to recognize it. Relatively, unsupervised learning has gathered less recognition in computer vision. We propose a Generative Adversarial Netwok(GAN) based model architecture, with a decoupled loss, trained in an unsupervised mode to reconstruct the corrupted regions of an image. We exercise a joint loss function comprising a reconstruction loss, an adversarial loss and a parsing loss to optimize our model. Our model is evaluated on Google Street View dataset. Our qualitative, as well as quantitative results, verify the state-of-the-art performance of our work in comparison to previous approaches. |
Year | DOI | Venue |
---|---|---|
2018 | 10.1109/ICACCI.2018.8554479 | 2018 INTERNATIONAL CONFERENCE ON ADVANCES IN COMPUTING, COMMUNICATIONS AND INFORMATICS (ICACCI) |
Keywords | Field | DocType |
Image Inpainting, Image Completion, Generative Adversarial Network, Convolutional Neural Network | Iterative reconstruction,Task analysis,Computer science,Control engineering,Inpainting,Supervised learning,Unsupervised learning,Artificial intelligence,Visual learning,Parsing,Decoding methods,Machine learning | Conference |
Citations | PageRank | References |
0 | 0.34 | 0 |
Authors | ||
3 |
Name | Order | Citations | PageRank |
---|---|---|---|
Nishat Afreen | 1 | 0 | 0.68 |
Shrey Singh | 2 | 5 | 1.10 |
Sanjay Kumar | 3 | 9 | 7.60 |