Suvobrat Ghosh

Amherst, MA 01002

I am a CS graduate @ UMass Amherst. I have over 5 years of work/research experience in computer vision, deep learning and software engineering. Currently, I am pursuing research in collaboration with Microsoft on lighweight architecture search (NAS) for speech recognition.

In the past, I have applied machine learning in a variety of domains such as finance, social media, telecommuncations, and education at organizations like Ericsson, PayTM CreditMate and IIT Bombay. I am polyglot in terms of languages, and like to explore various technology/solutions that might help in a scenario.

In general, I am really passionate about computer vision and machine learning, with a special interest in applied machine learning in vision problems such as autonomy, pathology, 3D vision etc. Check out my work on Github, or hit me up on LinkedIn if you’d like to connect!

posts

Jan 8, 2020	Backpropogation in batch normalization...
Jun 27, 2019	An introduction to neural networks

publications

IEEE

Determination of Porosity of Rock Samples from Photomicrographs Using Image Analysis

Debabrata Datta, Nikhil Thakur, Suvobrat Ghosh, and 2 more authors

In 2016 IEEE 6th International Conference on Advanced Computing (IACC), 2016

Abs HTML

Reservoirs are the porous and permeable rocks that contain commercial deposits of hydrocarbons. Porosity is a very important property of reservoirs. This paper presents a method of determining the porosities of different types of rocks based on image analysis. Stereological research for analysis of porosity were conducted by traditional methods before image analysis which were time consuming and lacked accuracy. The method proposed determines the porosity by computing the part of the whole sample for which the pores account. The steps involved in the above method are a series of contextual, non-contextual and morphological operations that are commonly used in image processing and analysis. The procedure was tested on thin sections of sandstone and limestone rock samples. The results were computed in the form of total porosity which includes all types porosities observed in rocks including isolated and connected porosities. The porosity obtained can also be called as visual porosity as it is being determined from photomicrographs. Values obtained show that the method proposed can lead to satisfying results. Obtained porosity values can be used further to determine determine other important properties of reservoir like permeability.