Manish Dhakal

Khumaltar, Lalitpur,

Bagmati, Nepal

Computer Engineer & Machine Learning Researcher

Hello, I’m Manish Dhakal, a Computer Engineering graduate from the Institute of Engineering, Pulchowk Campus, Tribhuvan University. Currently, I am working as a Research Assistant (RA) at NAAMII, specializing in segmentation and object detection tasks within the field of medical imaging.

My research projects encompass both unimodal and multimodal (esp. vision-language ) learning, for the purpose of segmentation and object detection in medical imaging. I have also been involved in a project of the segmentation task of 3D mesh data. My passion lies in exploring the potential of multimodal learning techniques.

Prior to my current role, I gained valuable experience as an intern in blockchain engineering and also worked as a fullstack developer and co-founder at a Nepal-based startup, Techniti Nepal. These experiences broadened my skill set and provided insights into various areas of technology and entrepreneurship.

During my undergraduate studies, our academic team worked on capstone project of automatic speech recognition for the low-resourced Nepali Language which was later presented at an IEEE conference.

I’m currently looking for new ML/AI research opportunities. My desire to contribute to these sectors and investigate cutting-edge solutions is motivated by a sincere curiosity. I’m looking forward to relevant Ph.D. programs to further my education and give back to the ML/AI community.

Selected Publications

MICCAI
Synthetic Boost: Leveraging Synthetic Data for Enhanced Vision-Language Segmentation in Echocardiography

Rabin Adhikari*, Manish Dhakal*, Safal Thapaliya*, and 3 more authors

arXiv preprint arXiv:2309.12829, 2023

Abs arXiv Bib PDF Code

Accurate segmentation is essential for echocardiography-based assessment of cardiovascular diseases (CVDs). However, the variability among sonographers and the inherent challenges of ultrasound images hinder precise segmentation. By leveraging the joint representation of image and text modalities, Vision-Language Segmentation Models (VLSMs) can incorporate rich contextual information, potentially aiding in accurate and explainable segmentation. However, the lack of readily available data in echocardiography hampers the training of VLSMs. In this study, we explore using synthetic datasets from Semantic Diffusion Models (SDMs) to enhance VLSMs for echocardiography segmentation. We evaluate results for two popular VLSMs (CLIPSeg and CRIS) using seven different kinds of language prompts derived from several attributes, automatically extracted from echocardiography images, segmentation masks, and their metadata. Our results show improved metrics and faster convergence when pretraining VLSMs on SDM-generated synthetic images before finetuning on real images. The code and prompts are released at https://github.com/naamiinepal/synthetic-boost
@article{adhikari2023synthetic, title = {Synthetic Boost: Leveraging Synthetic Data for Enhanced Vision-Language Segmentation in Echocardiography}, author = {Adhikari*, Rabin and Dhakal*, Manish and Thapaliya*, Safal and Poudel, Kanchan and Bhandari, Prasiddha and Khanal, Bishesh}, journal = {arXiv preprint arXiv:2309.12829}, year = {2023}, eprint = {2309.12829}, archiveprefix = {arXiv}, primaryclass = {cs.CV}, doi = {https://doi.org/10.48550/arXiv.2309.12829}, }