my pic

Kshitiz Garg

Former PHD Student

Vision Lab
Computer Science Department
Columbia University
Office: 616 CEPSR
Phone: 212-939-7090
Fax: 212-666-0140

My research

I graduated on May of 2007 and am currently working in a startup company in Boston area. My earlier research (during doctoral days) focused on improving outdoor vision in bad weather such as rain and snow. Rain and snow create sharp intensity fluctuations in images and videos, which degrade the performance of many outdoor vision algorithms such as feature tracking, segmentation, object recognition and tracking. My thesis work focused on handing rain. The main goal was to understand the visual effects produced by rain and develop models that lead to simple and effective algorithms for handling rain in images and videos and for its photorealistic rendering in graphics.

Recent publications

"Material Based Splashing of Water Drops," (EGSR 2007)

Splashing of water drops is one of the visually fascinating phenomena and capturing their interaction with scene elements is important in achieving realism. Splashing results from complex interactions between the drop and the material it impacts, which makes it hard to model analytically. This work takes an empirical approach. We measure the splashing behaviors of 22 common real word material and use it to develop a stochastic model for splash distribution. Our model is built upon empirical models previously developed in fluid dynamics and meteorology and only requires few parameters for generating splashes. User can tune the parameters to create novel splashes or combine different materials to generate physically plausible splashes for novel materials. The model is applicable for rendering splashes due to rain as well as water drops falling from large heights such as windowsills, trees, and rooftops.
[PDF[Project Page]

rain rendering thumbnail
"Photorealistic Rendering of Rain Streaks," (SIGGRAPH 2006)
Photorealistic rendering of rain streaks with lighting and viewpoint effects is a challenging problem. Raindrops undergo rapid shape distortions as they fall, a phenomenon referred to as oscillations. The interaction of light with the oscillating raindrops produces complex brightness patterns within a single motion-blurred rain streak. In this project, we develop a model for rain streak appearance that captures these complex interactions. Using this model we have developed a rain streak appearance database that contains thousands of rain streaks with different oscillation parameters and viewing and lighting directions.We have developed an efficient image-based rendering algorithm that uses our streak database to add rain to a single image or a captured video with moving objects and sources. Our rendering results show that the proposed physically-based rain streak model greatly enhances the visual realism of rendered rain.
[PDF[Project Page]                                                [Implemented in Microsoft SDK 10] (search topic rain )

"When Does a Camera See Rain?," (ICCV 2005)
The visibility of rain depends on various factors, such as, the camera parameters, the properties of rain and the brightness of the scene. Our analysis shows that the properties of rain – its small drop size, high velocity and low density – make its visibility strongly dependent on camera parameters such as exposure time and depth of field. In this project, we demonstrate how these parameters can be selected so as to reduce or even remove the effects of rain without altering the appearance of the scene. The proposed method serves to make vision algorithms more robust to rain without any need for post-processing.
[PDF[Project Page]

"Detection and Removal of Rain from Videos," (CVPR 2004)
In this project we conduct a comprehensive analysis of the visual effects of rain on an imaging system. This analysis includes a correlation model that captures the dynamics of rain and a physics-based motion blur model that captures the photometry of rain. Based on these models, we have developed simple and efficient algorithms for detecting and removing rain from videos. The effectiveness of our algorithms is demonstrated via experiments on videos of complex scenes with moving objects and time-varying textures. The techniques presented here can be used in a wide range of applications including video surveillance, vision based navigation, video/movie editing and video indexing/retrieval.
[PDF] [Project Page]


"Appearance of a Raindrop,"
(Tech. Report 2004)
Raindrops refracts and reflects both scene radiance and environmental illumination towards an observer. We have developed geometric and photometric models for refraction through, and reflection (both specular and internal) from, a rain drop. Our geometric and photometric models show that each rain drop behaves like a wide-angle lens that redirects light from a large field of view of 165 degrees towards the observer. Hence, each raindrop produces a wide angle view of the environment thus acting as a natural omni-directional imaging system. Our models provide important optical properties of raindrop which are useful for analyzing the appearance of rain.
[PDF] [demo]