Research

We train a robust CNN for White Blood Cell (WBC) classification by addressing cross-domain data imbalance and domain shifts.

We rethink the CNN classifier training in Raw image space instead of traditional RGB color space, diminish the illumination and coloring effects due to translate into the raw linear color space, and thus improve the classifiation performance.

Training the GAN-based SRResCSinGAN with learnable adaptive sinusoidal nonlinearities incorporated in LR and SR models by directly learn degradation distributions and then synthesize paired LR/HR training data to train the generalized SR model to real image degradations.

Training the RBSRICNN for Raw Burst Super-Resolution task that restore the clean RGB SR image by following the burst photography pipeline as a whole by a forward (physical) model.

Training the SR^2*GAN in a StarGAN like network topology with a single model to super-resolves the LR images for the multiple LR domains.

Training the BSRICNN for Burst Super-Resolution that solves the Burst SR task by iterative refinement of the intermediate SR estimates.

Training the ISRResDNet for Efficient Super-Resolution that solves the SR task as a sub-solver of the image denoising by the residual denoiser networks.

Training the SRResCycGAN network for Real-Image Super-Resolution with a generative adversarial network (GAN) framework for the LR to HR domain translation in an end-to-end manner.

Training the ISRResCNet network in an iterative manner with a residual learning approach for super-resolution to follow the image observation (physical) model.

Training the SRResCGAN network for super-resolution to follow the real-world degradation settings by adversarial training the model with pixel-wise supervision in the HR domain from its generated LR counterpart.

Training the SRWDNet for super-resolution that works for blur kernels of different sizes and different noise levels in an unified residual CNN-based denoiser network.

Deep Web Extractor system is a cloud-based web application that uses machine learning techniques for crawling and data discovery from the Deep Web (i.e., massive and quality portion of World Wide Web) to build knowledge based databases.