PhD Candidate · Kansas State University · Computer Science

Teaching machines
to see sport.

I'm Ahsan Zaidi — building computer vision systems that perceive and reason about human motion in real-world sports footage. My AutoTackle research pipeline detects risky tackles in football practice videos, with zero annotation cost and no task-specific training.

See the Research → GitHub LinkedIn Scholar Email
13Citations
2h-index
737Annotated Videos
CVPR2026

about

I'm a PhD candidate in Computer Science at Kansas State University, working on video understanding and spatiotemporal localization in sports contexts.

My AutoTackle research system automatically detects risky tackles in football practice videos — addressing a real problem where 2/3 of football head injuries happen during practice, and coaches currently review footage manually in a slow, subjective process that cannot scale.

My GRAZE pipeline runs completely training-free using GroundingDINO + SAM2 + motion-aware backward refinement to find the exact First Point of Contact frame. No frame-level labels, no fine-tuning, no domain data needed.

Stack

PythonPyTorchOpenCV SAM2GroundingDINOViViT I3DHuggingFaceCUDA SLURMLinuxHydra
Ahsan Zaidi

Citations / year

13Total
2h-index
2021
1
2022
2
2023
4
2024
6
2025
2
2026

visual research

GRAZE pipeline: Multi-frame grounding, backward refinement, SAM2 propagation to FPOC
GRAZE 4-phase pipeline  —  ① Multi-frame grounding  ·  ② Backward refinement to FFBO  ·  ③ SAM2 mask propagation  ·  ④ FPOC at first mask overlap
GRAZE SAM2 segmentation outputs: raw frames, dummy mask, player mask, segmented frame
GRAZE outputs — raw frame, dummy mask (SAM2), player mask, segmented overlay
GRAZE performance vs baselines: SOLE, MARS, TRACE, GRAZE (ours)
GRAZE vs baselines (SOLE, MARS, TRACE) — segmentation coverage, end-to-end coverage, conditional precision
Pose skeleton
Body mesh
3D side view
Pose estimation — skeleton overlay, SMPL body mesh, and 3D side reconstruction on live practice footage
TackleNet dataset sample footage
TackleNet dataset — 737 annotated football practice tackle clips, SATT-3 Strike Zone labels
Taguchi L18 augmentation search performance heatmap
ViViT Taguchi L18 augmentation search — performance heatmap across 18 runs (Run⊂15 = best: 66.7% risky recall)

projects

GRAZE pipeline
CVSports @ CVPR 2026

GRAZE — Zero-Shot Event Localization

Training-free pipeline for First Point of Contact (FPOC) detection in American football tackle videos. Combines GroundingDINO open-vocabulary grounding, SAM2 promptable segmentation, and motion-aware backward temporal refinement. Evaluated on 738 annotated clips with zero annotation cost.

PythonSAM2GroundingDINO Zero-ShotSLURMMIT License
View on GitHub →
Taguchi ViViT performance heatmap

Video Transformers · Action Classification

Tackle Study — ViViT & I3D

ViViT-B 16×2 and I3D fine-tuning for risky tackle classification. Tackles class imbalance (64.9% safe vs 35.1% risky) using Taguchi L18 augmentation search — 54 possible aug combos reduced to 18 systematic runs. Best config achieves 66.7% risky recall. Accepted at ICPR.

ViViTI3D PyTorchTaguchiFocal Loss
View on GitHub →

In Progress

More projects

Follow on GitHub for updates.

4.0
CGPA
7+
Disciplines
6
Publications
7
Courses Taught

multidisciplinary research

Computer Vision is the primary focus, but the research connects across a wide range of domains — from IoT systems and mechatronics to NLP, reinforcement learning, and network security. Each area informs the others.

👁 Computer
Vision Core
🤖Machine
Learning
🎮Reinforcement
Learning
💬NLP
📡IoT
⚙️Mechatronics
⚡Electrical
Eng.
🔒SDN
Security
🏈Sports
Analytics

teaching experience

Multidisciplinary teaching spanning Computer Science, Electrical & Electronics Engineering, and Mechatronics — from low-level logic circuits to high-level machine learning pipelines.

── Computer Science

🐍
Python Programming
Fundamentals through applied ML — data structures, NumPy, pandas, PyTorch, and computer vision pipelines.
⌨️
C Language
Memory management, pointers, data structures, and systems-level programming for embedded and real-time applications.

── Electrical Engineering & Mechatronics

🔢
Digital Logic Design
Boolean algebra, combinational & sequential circuits, FSMs, and hardware description fundamentals.
⚡
Electric Circuits
DC/AC analysis, Kirchhoff's laws, Thevenin/Norton theorems, phasors, and frequency response.
⚙️
Engineering Dynamics
Kinematics and kinetics of particles and rigid bodies — Newton-Euler formulations with robotic system applications.
📊
Engineering Statistics
Probability distributions, hypothesis testing, regression, and statistical inference for engineering design.

── Tools & Simulation

📉
MATLAB
Numerical computing, signal processing, Simulink-based system modelling, and control system design.
Spanning: Computer Science Electrical Engineering Mechatronics Applied Mathematics Control Systems ML & Data Science

papers

2026 CVSports @ CVPR 2026

GRAZE: Grounded Refinement and Motion-Aware Zero-Shot Event Localization

Ahsan Zaidi, Lior Shamir, William Hsu, Scott Dietrich, Talha Zaidi · arXiv:2604.01383
2025 ICPR · arXiv 2025

ViTs for Action Classification in Videos: An Approach to Risky Tackle Detection in American Football Practice Videos

Syed Ahsan Masud Zaidi, William Hsu, Scott Dietrich · arXiv:2604.01318
2025 SSRN Preprint

Advancing Remote and Continuous Cardiovascular Patient Monitoring through a Novel IoT-Driven Framework

S. Nayab, ..., SAM Zaidi, et al. · arXiv:2505.03409
Full profile on Google Scholar →

contact

Let’s connect.

Open to research collaborations, sports analytics problems, and conversations about video understanding, zero-shot CV, and player safety technology. Based in Manhattan, KS.

ahsanzaidi@ksu.edu GitHub LinkedIn Scholar arXiv