Naveed Akhtar

Language-dependent biases in text-to-image generation reveal significant gaps in model performance across different cultures and languages.

Introspective attention modulation can significantly enhance the safety of T2I models without sacrificing quality, outperforming existing methods like concept erasure.

Existing image comparison methods miss the mark in UI testing, but a new benchmark reveals that advanced techniques can drastically cut down on irrelevant noise.

Injecting intermediate text representations into the denoising process can dramatically enhance the alignment of text-to-image models with challenging prompts for unique objects.

ReasonCLIP-58M shows that structured reasoning supervision can dramatically boost CLIP's reasoning abilities while maintaining efficiency.

Label-preserving self-saliency mixup can significantly boost model performance while maintaining semantic integrity in data augmentation.

Achieving state-of-the-art results in pose estimation, this method leverages symmetry-aware features to simplify the complexities of rotation learning in a non-linear space.

Brain tumor segmentation can now maintain accuracy even when crucial MRI data is missing, thanks to a novel diffusion-based imputation strategy.

Retraining just the classifier head of a frozen feature extractor can be dramatically improved by meta-learning feature-space augmentations that target hard examples, leading to state-of-the-art robustness against spurious correlations.