Kishan Sharma

• Applied methods developed during master thesis on Breast Cancer Dataset to predict count and size histogram of malignant cancer cells • Fine-tuned our model using object count and size histogram as intermediated results and improved cellularity score prediction • Submitted a research paper, ”HistoNet-Predicting size histograms of object instances” in International Conference on Computer Vision 2019

• This thesis aims to develop complete solution of automation of most laborious part of Bio-waste conversion by counting and estimating size of insect larvae • Built a novel deep learning architecture (HistoNet), which counts and predicts the size distribution of objects directly from an input image, used deep supervision at hidden layers to further improve our method, showing superior results with respect to state-of-the-art instance based segmentation methods while having 85% less parameters

• Modeled and trained Deep Learning architecture for predicting tray motion, tray level and for classifying product from outside and inside of the oven for automatic baking program selection • Exploited temporal patterns with Deep Recurrent Neural Networks (GRU) for the estimation of thickness of the product, baking time estimation and product recognition using input video data and sensor data as in intermediate input

Computer Vision & Machine Learning, GPA - 1.5

Computational Fluid Dynamics, GPA-2.0