Skip to main content
Nauman Irshad Lab

Research area

Computer Vision

PoseDepth-CMP (OpenPose COCO vs MPI), Smart Fitao AI 3D try-on / size prediction, and crop-disease vision systems.

Projects
3
Publications
1
Conference
0
Datasets
2
Software
1

Overview

Computer vision at Nauman Irshad Lab covers person pose and monocular depth (PoseDepth-CMP), Metaverse / PIFuHD virtual try-on for Smart Fitao AI, and plant disease recognition with ResNet50 and Grad-CAM for Agri Asia 2026.

Student researchers co-author papers and ship demos presented at LCCI and Agri Asia.

Objectives

  1. Compare OpenPose body models for contour-guided person depth estimation.

  2. Deliver practical cloth size prediction and 3D try-on for tailors and sellers.

  3. Ship explainable plant-disease classifiers for agricultural exhibition and field use.

Future directions

  • Stronger Metaverse try-on fidelity with PIFuHD finetuning.
  • Mobile deployment of Grad-CAM-backed crop disease tools.

Research images

From the laboratory

Full gallery
Computer Vision — laboratory imagery
Computer Vision — laboratory imagery

Current projects

Projects in computer vision

All projects
PoseDepth-CMP · OpenPose COCO vs MPI for Person Depth Estimation project imagery
Under reviewComputer Vision2025 – 2026

PoseDepth-CMP · OpenPose COCO vs MPI for Person Depth Estimation

PoseDepth-CMP gives practitioners evidence for choosing between OpenPose body-model variants for person-specific monocular depth. On 240 controlled samples, COCO reaches higher keypoint accuracy and lower depth error; MPI is faster and lighter. The project contributes paired statistics, cue ablations, uncertainty analysis and a learned adaptive selector.

Supervisor & research team

Click a photo to open LinkedIn

Technology stack

  • Python
  • OpenPose
  • COCO
  • MPI
  • Contour analysis
  • NumPy

Paper · download & cite

Download PDF
Smart Fitao AI · Cloth Size Prediction, 3D Try-On & Seller Studio (FYP) project imagery
OngoingComputer Vision2024 – 2026

Smart Fitao AI · Cloth Size Prediction, 3D Try-On & Seller Studio (FYP)

Smart Fitao AI asks: what if getting the perfect fit required no guesswork — just AI? The Final Year Project delivers (1) cloth size prediction for tailors from body and garment measurements, (2) 3D virtual try-on for a more personalised shopping experience, (3) a 24/7 AI chatbot with 3D product recommendations, and (4) a 3D studio for sellers to convert 2D products into interactive 3D models (including PIFuHD Metaverse finetuning). Presented at the Lahore Chamber of Commerce and Industry (LCCI). FYP advisor: Asif Farooq. Team: Nauman Irshad Ali Shah, Umer Amir, Ali Ahmad and Abdul Rehman.

Publications
In preparation

Supervisor & research team

Click a photo to open LinkedIn

Technology stack

  • Python
  • PIFuHD
  • PyTorch
  • Flutter
  • Computer Vision
  • Metaverse
  • Chatbot
  • 3D Studio
Crop Disease Detection · ResNet50 + Explainable AI for Agricultural Intelligence project imagery
CompletedDeep Learning2025 – 2026

Crop Disease Detection · ResNet50 + Explainable AI for Agricultural Intelligence

This AgriTech project detects 15 plant diseases across tomato, potato and pepper bell using 20,638 images, transfer learning on ResNet50 (ImageNet) and Explainable AI via Grad-CAM for transparent predictions. Reported metrics: 90.09% accuracy, 88.41% precision, 89.18% recall and 88.73% F1, with sub-200ms FastAPI inference. Presented as an exhibitor at the 19th International Agri Asia & Green Pakistan Exhibition & Conference (09–11 May 2026, Expo Centre Lahore). Team: Nauman Irshad Ali Shah, Ali Ahmad, Abdul Rehman, Danish Ali and Umer Amir.

Publications
In preparation

Supervisor & research team

Click a photo to open LinkedIn

Technology stack

  • Python
  • ResNet50
  • PyTorch
  • Grad-CAM
  • FastAPI
  • Transfer Learning
  • Explainable AI

Publications

Peer-reviewed outputs in this area

Search all publications
First page of PoseDepth-CMP: A Comparative Analysis of OpenPose COCO and MPI Keypoint Models for Contour-Guided Monocular Person Depth Estimation with Adaptive Selection
Journal paperUnder review2026

PoseDepth-CMP: A Comparative Analysis of OpenPose COCO and MPI Keypoint Models for Contour-Guided Monocular Person Depth Estimation with Adaptive Selection

Nauman Irshad Ali Shah, Sammra Habib, Muhammad Umer Amir, Ali Ahmad, Danish Ali

Journal manuscript · Computer Vision / Pose & Depth

Read abstract

Monocular depth estimation from human imagery supports clinical gait analysis, sports biomechanics, surveillance, and human–computer interaction, yet practitioners lack direct evidence for selecting between OpenPose body-model variants. PoseDepth-CMP provides a locked-seed comparison of COCO (18 keypoints) and MPI (15 keypoints) within the same contour-guided geometric depth pipeline. On 240 controlled samples (seed 42), COCO achieves 94.4% keypoint accuracy with 13.0 cm depth error, whereas MPI achieves 87.3% accuracy with 15.4 cm error while reducing runtime by 13.3% and memory by 16.7%. The learned selector attains 93.6% accuracy and 12.4 cm depth error at intermediate computational cost.

Keywords: OpenPose · COCO keypoints · MPI keypoints · monocular depth estimation · human pose estimation · adaptive model selection

PDF
Computer VisionDeep Learningartificial-intelligencedata-science

Continue exploring

Other research areas