Proposed a state of the art segmentation network for classify lesion pixel in demroscopic images.
A two stage architecture first localizes the possible skin lesion and then are fed to the segmentation network.
Segmentation network was pretrained on the cropped dataset prepared from the original masks.
By :- Shreshth Saini & Divij Gupta
We propose a 3D architecture for the localization of the 3D data to produce tightly fit 3D volumetric samples.
A modified version of the Faster RCNN based architecture is developed for the localization and producing cubic samples.
The cubes are then fed to a 3D U-Net + Hourglass network for generating 3D masks
By :-Shreshth Saini & Divij Gupta
An end-to-end architecture to segment the human-Ear was designed in keras.
The architecture was inspired from UNet and was modified intensively and performed well.
Further work introduces the pix2pix GANs for the segmentation task.
By :-Shreshth Saini & Divij Gupta
This project focuses on investigation and development of an algorithms associated with recognition of Pathologic Myopia (PM) and classification of fundus images taken from patients suffered from PM disease into 3 labels:
Pathological Myopia
High Myopia
Myopia is one of the optical disease having highest morbidity. More than 2 billion people have myopia in whole world, among which 35% are suffering from high myopia, it is important to have early diagnosis and regular follow-up. This has motivated the research and development of Classification network so that it can be detected at early stage and can be cured.
By :-Anupama Patel
Our architecture had RES-NET50 as feature extractor and a Fully connected classifier classifying into 6 classes.
Made our own data-set consisting of 1050 images of various Document Embedded images of class Tables, Graphs, Charts, etc
By :-Mayank Maheshwari
The challenge have five tasks:-
Developed a segmentation network inspired from UNet having special blocks in encoder and decoder part, used novel training methodology, and achieved satisfying results
By :-Nisarg Shah
Proposed an architecture for localization of cancerous cells in images
Implemented a segmentation architecture followed by regression model for producing the results that beat earlier state-of-the-art results available
By :-Nisarg Shah
Major work includes segmentation of lung CT scan,detection of lung nodule candidates ,extraction of features and to classify nodule candidates into nodule or non-nodule.
Project includes the use of image processing operations like morphological operations and spatial filtering.
By :-Geetika Agrawal