"Muhammad Talha — 3DGS • Point Cloud Compression • 3D Point Cloud Deblocking/Denoising • AR/VR • ML Systems"
PhD @ UMKC • Ex-Intern Qualcomm R&D (’25) • 3D Gaussian Splatting • 3D Point Clouds
• Denoising/Deblocking • Graph Signal Processing • Compression
About · Research Focus · Research Papers · Patents & IP · Research Projects · Awards & Achievements · Tech Stack · Contact
|
|
- Temporal prediction for 3DGS: inter-Frame Prediction Framework for AR/VR applications.
- Intra-frame Coding for 3D Gaussian Splatting Sequences.
- Quantization: Vector Quantization, Learned Codebooks for efficient compresison of SH channel of 3DGS.
- Geometry & entropy: GPCC geometry, LZMA/7-Zip, per-group bit allocation.
- 3D Point Clouds: Denoising, Deblocking as a post-processing framework to remove coding induced artifacts.
- Graph Fourier Transform: Signal Processing, Singal projection in Frequency domain for efficient multi-scale deblocking framework.
Add DOIs/arXiv as they go live. Keep titles concise and outcome-oriented.
| Year | Venue | Title | Status | Links |
|---|---|---|---|---|
| 2026 | DCC | InterGS-Lite: Light Weight Dynamic GS coding with Vector Quantization of Prediction Residuals | Under Review | ⟦draft⟧ |
| 2026 | TIP | Channel-wise Transformer-based Hierarchical Feature Fusion for Point Cloud Attribute Compression Artifact Mitigation | Under Review | ⟦draft⟧ |
| 2025 | VCIP | InterGS: Inter-Frame Prediction for Dynamic 3D Gaussian Splatting | accepted | ⟦PDF⟧ ⟦Poster⟧ |
| 2025 | ICIP | J-SGFT: Joint spatial and graph fourier domain learning for point cloud attribute deblocking | accepted | ⟦PDF⟧ ⟦Poster⟧ |
| 2024 | DCC | MUSCON: Multi-scale SparseConv Learning for Point Deblocking | accepted | ⟦PDF⟧ ⟦Poster⟧ |
| 2024 | MMSP | Sparse Convolution Based Point Cloud Attributes Deblocking with Graph Fourier Latent Representation | accepted | ⟦PDF⟧ ⟦Poster⟧ |
| 2024 | Artificial Intelligence and Fuzzy Logic System 2024 | Yolov5, yolo-x, yolo-r, yolov7 performance comparison: A survey | accepted | ⟦PDF⟧ ⟦Poster⟧ |
Mini abstracts
-
InterGS (VCIP 2025). Predict current-frame Gaussian attributes from a reference I-frame; transmit residuals only. Dual predictor (bilateral + KNN) with per-Gaussian mode selection cuts bitrate while preserving photorealistic quality.
-
InterGS-VQ (DCC 2026, target). Vector-quantize SH_DC/SH_AC with feature-aware codebooks, keeping opacity/scale/rot scalar. Gains in BD-Rate with stable PSNR across multi-frame chains.
| ID / Ref | Title | Area | Status |
|---|---|---|---|
| ⟦To be Entered Later⟧ | InterGS: Dynamic Inter-Prediction for 3D Gaussian Splatting | 3DGS Inter-frame compression | Patent Pending |
Pin these three (Customize Profile → Pinned).
- Meta Hacker Cup 2025 participant (Round 1 solutions public - On-going).
- Qualcomm Multimedia R&D (’25): inter-prediction for 3DGS (patent-pending).
- Google Jam 2023: Highest ranked over Pakistani Leaderboard.
- Advent of Code 2022: Top 6th on Pakistani Leaderboard.
- Won Research Funding on National level from IT Ministry, Pakistan.
- Secured merit-based fully funded scholarship in Undergrad.
- Solved over 250 Coding questions on Leetcode.
- RD curves: BD-Rate/PSNR with matched rate points.
- Decode performance: ms/gaussian + FPS targets per sequence.
- Artifacts: quant configs, codebook dumps, predictor-choice streams, geometry bytes.
