Md. Mahbubur Rahman

Practical deep learning

Mon, 01 Jan 0001 00:00:00 +0000

A short written introduction to tensors, training, and evaluation. Video recordings and slides can be added to each lesson when available.

Tensors and shapes

Mon, 01 Jan 0001 00:00:00 +0000

A tensor stores values in a multidimensional array. In image learning, dimensions often describe the batch, channels, height, and width.

For a batch of eight RGB images resized to 224 by 224 pixels, the usual PyTorch shape is (8, 3, 224, 224). A single-channel depth input would normally have shape (8, 1, 224, 224).

import torch
rgb = torch.zeros(8, 3, 224, 224)
depth = torch.zeros(8, 1, 224, 224)
rgbd = torch.cat([rgb, depth], dim=1)
print(rgbd.shape) # torch.Size([8, 4, 224, 224])

Concatenation only aligns array dimensions. It does not register the cameras, convert depth units, or remove invalid depth values. Those steps must be handled before treating the channels as corresponding observations.

Try it

Change the batch size, then inspect every tensor shape. Consider which preprocessing steps are shared between RGB and depth and which require separate handling.

Training and evaluation

Mon, 01 Jan 0001 00:00:00 +0000

Training adjusts model parameters using examples and a loss function. Validation informs choices such as hyperparameters and stopping time. The test set evaluates the final procedure after those choices are fixed.

A reliable sequence

Define the prediction target and the unit of independence.
Split the data before fitting preprocessing or selecting a model.
Fit using training data and choose settings using validation data.
Evaluate the frozen procedure on the held-out test data.

For repeated observations of plants, an image-level random split can place the same plant in both training and test sets. A plant-level split is more appropriate when the goal is performance on unseen plants.

Evaluation mode

model.eval()
with torch.inference_mode():
 predictions = model(test_inputs)

Evaluation mode changes the behavior of layers such as dropout and batch normalization. Disabling gradient recording reduces the work required during inference. These operations serve different purposes.

Try it

Write down the grouping variable for your own dataset. Check whether any group appears in more than one partition.

Working with RGB-D observations

Mon, 01 Jan 0001 00:00:00 +0000

RGB describes appearance. Depth describes distance according to the sensor’s coordinate system and unit convention. Combining them requires attention to calibration, alignment, missing measurements, and scale.

Before modeling

Record the sensor, depth units, and calibration used for each collection.
Verify alignment between the color image, depth image, and any segmentation masks.
Keep track of invalid depth values instead of treating them as valid zero-distance observations.
Apply spatial transformations consistently across corresponding channels and masks.

Evaluating a benefit from depth

Compare RGB and RGB-D models using the same data partitions and training budget. Check results by occlusion severity or other relevant conditions. An improvement in aggregate performance alone does not explain when depth contributes.

Try it

Inspect a color image, a depth image, and a validity mask side by side. Look for boundary misalignment and regions where the sensor did not return a usable measurement.

CLFM: climate-informed representation learning

Wed, 16 Sep 2026 00:00:00 +0000

CLFM explores label-efficient plant trait prediction by bringing greenhouse images and climate measurements into a shared learning framework. Current development investigates image-environment contrastive learning and adaptation of vision backbones.

TAHAF: RGB-D plant phenotyping

Wed, 16 Sep 2026 00:00:00 +0000

TAHAF uses trait-adaptive hierarchical attention fusion to combine RGB and depth information for lettuce phenotyping in hydroponic systems.

Temporal amodal leaf completion

Wed, 16 Sep 2026 00:00:00 +0000

This project investigates recovery of occluded plant leaves using RGB-D observations and an identity-matched history across days. The methodology separates mask reconstruction from reliability estimation. Work includes correspondence auditing, independent evaluation, and controlled ablations.

Attention-Based Task-Adaptive Fusion for Real-Time RGB-D Lettuce Morphological Analysis

Wed, 01 Jul 2026 00:00:00 +0000

Presented at the ASABE Annual International Meeting in Indianapolis in July 2026.

TAHAF: Trait Adaptive Hierarchical Attention Fusion of RGB-D data for automated lettuce phenotyping in hydroponic systems

Mon, 01 Jun 2026 00:00:00 +0000

FruitsMultiNet: Deep neural network for fruit identification through multi-scale feature fusion

Sun, 01 Jun 2025 00:00:00 +0000

FruVeg_MultiNet: A hybrid deep learning-enabled IoT system for fruit and vegetable identification

Sat, 01 Mar 2025 00:00:00 +0000

Smart farming: Leveraging IoT and deep learning for sustainable tomato cultivation and pest management

Fri, 01 Nov 2024 00:00:00 +0000

AirNet: Predictive machine learning model for air quality forecasting

Tue, 01 Oct 2024 00:00:00 +0000

Hybrid feature fusion and CNN for melanoma skin cancer detection

Sun, 01 Oct 2023 00:00:00 +0000

BreastMultiNet: Multi-scale feature fusion for breast cancer detection

Thu, 01 Dec 2022 00:00:00 +0000

Academic biography

Mon, 01 Jan 0001 00:00:00 +0000

I am Md. Mahbubur Rahman, a PhD student in Computer Science and Graduate Research Assistant in the Department of Electrical Engineering and Computer Science at the University of Wyoming. I work with Dr. Yaqoob Majeed in the AgBodied Lab.

My research focuses on computer vision and deep learning for plant phenotyping in controlled environments. I use RGB-D imaging to study plant structure and estimate growth traits without destructive sampling.

I also serve as Research Director at in Bangladesh. Before starting my doctoral studies, I held teaching and academic leadership roles at East West University, Dhaka International University, and Bangladesh University of Business and Technology.

This website brings together my research, teaching resources, and writing. My journal includes research notes, practical tutorials, reflections on doctoral study, and selected moments from family life.

or .

Accessibility

Mon, 01 Jan 0001 00:00:00 +0000

This website aims to support keyboard navigation, readable text, mobile layouts, and meaningful headings. If you encounter a barrier, please email with the page address and a description of the problem.

Contact

Mon, 01 Jan 0001 00:00:00 +0000

I welcome conversations about research collaboration, teaching, mentoring, and shared interests in computer vision and agricultural AI.

mrahma11@uwyo.edu

LinkedIn | Google Scholar | ORCID

GitHub

Laramie, Wyoming, USA.

CV and experience

Mon, 01 Jan 0001 00:00:00 +0000

Education

PhD in Computer Science, University of Wyoming, January 2025 to present.
MSc in Computer Science and Engineering, MBSTU, 2016 to 2020.
BSc in Computer Science and Engineering, MBSTU, 2012 to 2016.

Research and academic appointments

Graduate Research Assistant, AgBodied Lab, University of Wyoming, January 2025 to present.
Research Director, Peer Research Lab, Bangladesh.
Assistant Professor, Bangladesh University of Business and Technology, August 2023 to March 2025.
Assistant Professor and Acting Chairman, Dhaka International University, January 2022 to July 2023.
Lecturer, East West University, September 2017 to December 2021.

Industry experience

Chief Executive Officer, Bemantech Ltd., Bangladesh, March 2023 to February 2025.

Technical experience

Python, C, C++, Java, PyTorch, TensorFlow, OpenCV, Open3D, RGB-D imaging, multimodal sensor fusion, Git, LaTeX, and GPU computing with SLURM.

and .

Leadership and service

Mon, 01 Jan 0001 00:00:00 +0000

Research leadership

As Research Director at Peer Research Lab, I support collaborative research and mentoring across applied AI topics.

Teaching and supervision

My academic work in Bangladesh included university teaching, academic coordination, and thesis supervision. My CV records supervision of more than 40 undergraduate and graduate thesis projects across these appointments.

University service

Former Vice President, ACM Student Chapter, University of Wyoming.
Former Vice President, International Students Association, University of Wyoming.

Peer review

I have reviewed for journals including Computers and Electronics in Agriculture and Informatics in Medicine Unlocked.

Peer Research Lab

Mon, 01 Jan 0001 00:00:00 +0000

I serve as Research Director at Peer Research Lab in Bangladesh. My interests include collaborative research, student mentoring, and developing rigorous applied AI studies.

Research and collaboration

My work connects computer vision and deep learning with questions in agriculture and other applied domains. I welcome conversations about shared research interests, evaluation methods, and student projects.

Mentoring

If you are interested in discussing a research direction, include your background, the question you want to investigate, and any relevant prior work in your message.

or .

Privacy

Mon, 01 Jan 0001 00:00:00 +0000

This site publishes research information, educational materials, and selected personal writing. The site itself does not use advertising trackers or a comment system.

The hosting provider may process request information needed to operate and secure the website. Following an external link takes you to a service governed by its own privacy practices. Embedded video services, when used, may also process information about your visit.

The editing dashboard uses GitHub authentication for authorized editors. Visitors do not need an account to read public pages.

For questions about this website, email .

Research

Mon, 01 Jan 0001 00:00:00 +0000

My current research sits at the intersection of computer vision, multimodal learning, and controlled-environment agriculture. I focus on using observations that can be collected repeatedly without destroying the plant.

RGB-D plant phenotyping

studies attention-based fusion of color and depth for estimating lettuce traits.

Temporal amodal leaf completion

investigates how observations of the same leaf across days can inform recovery of regions hidden in the current frame. Identity correspondence, evaluation quality, and reliability estimation are central to this work.

Climate-informed representation learning

explores image and environmental information for learning useful representations with fewer trait labels.

Longer-term direction

I am interested in connecting perception and growth models to greenhouse digital twins and decision support.

or .

Resources

Mon, 01 Jan 0001 00:00:00 +0000

Programming and deep learning

Computer vision and 3D data

Research practice

Teaching and lectures

Mon, 01 Jan 0001 00:00:00 +0000

My teaching background spans university appointments in computer science in Bangladesh. Here I share written tutorials and resources connected to programming, computer vision, and deep learning.

Lecture library

The organizes materials by topic. Each series can include written lessons, video recordings, slides, notebooks, and further reading.

Learning resources

The links to useful documentation and tools. Materials will grow as I prepare and publish them.