|
Annapoorna Sai Sriram Mandalika
I am a final-year Undergraduate student at SRM Institute of Science and Technology, Chennai advised by Dr. Athira M. Nambiar, where my focus is vision model agnostics, leaning methodologies and reasoning.
I enjoy traveling and spending time with loved ones, and I'm open to discussing research or CV system development opportunities—feel free to email me.
Email /
CV /
LinkedIn
|
|
|
Research
My research interests mainly lie in the areas of computer vision and deep learning. Partiuclarly focused on learning methods for vision models, label-efficient (Semi-Supervised/ Unsupervised /Self-Supervised ) approaches and building agents that can independently explores environment and act in the world by learning from data and self-intuitive interactions
In addition, I am also interested in classical computer vision areas such as representation learning ,domain adaptation and transfer learning.
Google Scholar /
Github
|
I’ve worked with some amazing people, especially (in no particular order) Dr. Athira Nambiar, Prof. C. Krishna Mohan, Mr. Mrinmay Sen, Mr. Rahul Biswas, Ms. Sai Veena Suresh, Mr. Sampath Kumar, and Ms. Shilpi Garg for supporting my research journey.
Highlights
- Paper on Precognitive Uncertainty-aware Chain-of-Thought (CoT) for driving scene scenario accepted at CVPR 2025 , arxiv (soon)
- Paper on Semi-Supervised semantic segmentation for driving scene scenario accepted at ICPR 2024 , arxiv
- Book chapter on Speech emotion analysis IGI Global
- Student research collaborator at IIT Hyderabad, explored Continual Learning and Image Reconstruction methods.
|
For the up-to-date publication list, please visit the Google Scholar page.
* Equal contribution. † Equal advising.
|
PRIMEDrive-CoT: A Precognitive Chain-of-Thought Framework for Uncertainty-Aware Object Interaction in Driving Scene Scenario
Sriram Mandalika , Lalitha V and Athira Nambiar
IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025
In this work, we introduce PRIMEDrive-CoT, an uncertainty-aware model for object interaction and Chain-of-Thought (CoT) reasoning in driving scenarios. It integrates LiDAR-based 3D detection with multi-view RGB for interpretable scene understanding. Bayesian GNNs model uncertainty and interactions, while CoT reasoning and Grad-CAM enhance interpretability. Evaluated on DriveCoT, PRIMEDrive-CoT outperforms existing CoT and risk-aware models.
|
|
SegXAL: Explainable Active Learning for Semantic Segmentation in Driving Scene Scenarios
Sriram Mandalika and Athira Nambiar
27th International Conference on Pattern Recognition (ICPR), 2024
In this work, we propose a novel active learning mechanism for semantic segmentation for self-driving vehicles using the Cityscapes dataset. We leverage the iterative data-pipeline to improve the model accuracy using the least amount labelled images along with a blend of XAI and entropy analysis.
|
|
Deep Learning-Based Speech Emotional Analysis Using Convolution Neural Network: Bi-Directional Long Short-Term Memory
S Aruna, G Usha, A Saranya, M Maheswari, MASS Mandalika
IGI Global, 2024
This book chapter, titled "Machine and Deep Learning Techniques for Emotion Detection," delves into the application of deep learning (DL) as part of artificial intelligence (AI) to perform tasks requiring human intelligence. Specifically, it focuses on speech emotion recognition (SpEmRe), which identifies various emotions from audio samples. DL techniques, especially bi-directional long short-term memory (Bi-LSTM) models, are utilized to predict human emotions from speech, reflecting the growing interest.. Read more
|
Services
|
-
Website template from here
* denotes equal contribution
|
|