Publications

Refer to my Google Scholar page for a more up-to-date list of publications.

NoisyTwins: Class-Consistent and Diverse Image Generation through StyleGANs

NoisyTwins is a self-supervised regularization scheme for StyleGANs, which helps in alleviating mode collapse and leads to consistent conditional fine-grained image generation.

H. Rangwani, L. Bansal, K. Sharma, T. Karmali, V. Jampani, R.V. Babu

Conference on Computer Vision and Pattern Recognition, CVPR’23

Project Page / PDF / Code

Hierarchical Semantic Regularization of Latent Spaces in StyleGANs

Hierarchial Regularization of GAN features which enhances the smoothness in latent space

T. Karmali, R. Parihar, S. Agarwal, H. Rangwani, V. Jampani, M. Singh, R.V. Babu

European Conference on Computer Vision, ECCV’22

Project Page / PDF / Code

Improving GANs for Long-Tailed Data through Group Spectral Regularization

Group Spectral Regularization for alleviating mode collapse in GANs particulary for long-tailed data.

H. Rangwani, N. Jaswani, T. Karmali, V. Jampani, R.V. Babu

European Conference on Computer Vision, ECCV’22

Project Page / PDF / Code

Everything is There in Latent Space: Attribute Editing and Attribute Style Manipulation by StyleGAN Latent Space Exploration

Technique to find high quality GAN edit direction using very few images.

R. Parihar, A. Dhiman, T. Karmali, R.V. Babu

ACM International Conference on Multimedia, ACMMM’22

Project Page / PDF

LEAD: Self-Supervised Landmark Estimation by Aligning Distributions of Feature Similarity

Reported correspondence tracking capability in BYOL and utilized it for few-shot landmark estimation.

T. Karmali*, A. Atrishi*, S.S. Harsha, S. Agarwal, V. Jampani, R.V. Babu

Winter Conference on Applications of Computer Vision, WACV’22

Project Page / PDF / Code

Deep Implicit Surface Point Prediction Networks

3D Shape representation for both open and closed shapes, also enables fast surface normal finding.

R.M. Venkatesh, T. Karmali, S. Sharma, A. Ghosh, R.V. Babu, L.A. Jeni, M. Singh

International Conference on Computer Vision, ICCV’21

Project Page / PDF / Code

 

Masters Thesis

Landmark Estimation and Image Synthesis Guidance using Self-Supervised Networks

Discovered properties neural networks trained on large data using Self-Supervised Learning, and used them to improve few-shot landmark estimation and naturalness of generated images in StyleGANs.

T. Karmali

M.Tech. (Research) Thesis, Indian Institute of Science, March, 2022

pdf