Publications
- Evaluating 3D Shape Analysis Methods for Robustness to Rotation Invariance
- Accepted for Oral Presentation at Conference on Robots and Vision (CRV) 2023
- We analyze the robustness of recent 3D shape descriptors to SO(3) rotations, something that is fundamental to shape modeling.
- We benchmark different methods for feature extraction and classification in the context of Instance Classification task.
- We systematically contrast different choices in a variety of experimental settings investigating the impact on the performance under:
- different rotation distributions
- different degrees of partial observations on the object
- different levels of difficulty of negative pairs
- Our study provides useful pointers in making design choices for modeling rotation-invariance for shape encoding.
- Project website, Link to paper, Cite(BibTex)
STWalk: Learning Trajectory Representations in Temporal Graphs
Accepted as main research track paper in CoDS-COMAD 2018. STWalk analyzes the temporal behavior of nodes in dynamic graphs by doing random walks on graph at a current & past time-steps. STWalk outperforms baseline algorithms on 3 real-world datasets. Implemented in Python, TensorFlow-Keras.
Survey of Recent Advances in Visual Question Answering
The paper describes the approaches taken by various algorithms to extract image features, text features and the way these are employed to predict answers.
Community-based Outlier Detection in Edge-attributed Graphs
The paper introduces a novel method that detects holistic outlier graph nodes by taking into account the node data and edge data simultaneously to detect anomalies.