Deep Learning & Computer Vision Projects

A curated collection of production-style deep learning projects in PyTorch and TensorFlow/Keras — covering computer vision, generative models, adversarial learning, NLP, and time-series analysis. Every project is a self-contained, modular package with a CLI, configuration, tests where applicable, and its own documentation.

About

A collection of production-ready deep learning projects in PyTorch and TensorFlow — vision transformers, GANs, adversarial attacks, face recognition, image captioning, pose estimation, semantic segmentation, NLP topic modeling, and video analysis. Each project ships as a clean, modular package with CLI entry points, reproducible training, and full documentation.

GitHub Topics

Paste into the repo About → Topics field:

deep-learning, machine-learning, computer-vision, pytorch, tensorflow,
keras, vision-transformer, transformers, gan, generative-models,
adversarial-attacks, image-captioning, pose-estimation, image-segmentation,
face-recognition, object-detection, yolo, nlp, topic-modeling,
sentiment-analysis, time-series, jupyter-notebook, cnn, lstm, opencv

Projects

Every folder is an independent project with its own README.md, code, and requirements. The original Jupyter notebooks are kept untouched as the research record.

Computer Vision

Project	What it does	Stack
vision-transformer	From-scratch ViT & CCT vision transformers trained on CIFAR-10/100	PyTorch
video-YOLO	Driving-video analysis: lane detection + YOLOv5 vehicle detection	PyTorch, OpenCV
pose estimation	Human pose estimation via heatmap regression (LSP)	PyTorch
U-NET	Binary image segmentation with a pretrained U-Net	TensorFlow/Keras
image-color-clustring	K-Means color quantization and palette extraction	OpenCV, scikit-learn
STN-and-transferLearning	Spatial transformer networks + transfer learning (ResNet, DeiT)	PyTorch, timm

Generative & Adversarial

Project	What it does	Stack
GAN	GAN & DCGAN trained on Fashion-MNIST / CelebA	TensorFlow/Keras + PyTorch
adversarial	FGSM & DeepFool adversarial attacks with accuracy analysis	PyTorch

Faces & Biometrics

Project	What it does	Stack
siameseNet	Face recognition with a Siamese network + triplet loss (LFW)	PyTorch
FER	Facial expression recognition across 7 emotions (FER2013)	PyTorch

Vision + Language

Project	What it does	Stack
image_captioning	CNN-LSTM image captioning (Inception-v3 + LSTM, Flickr8k)	PyTorch

NLP

Project	What it does	Stack
topic-modeling (NLP)	Unsupervised topic discovery in Persian news (embeddings + DBSCAN/community detection)	sentence-transformers, scikit-learn
_ML/sentiment_analysis_IMDB	IMDB sentiment analysis with TF-IDF + classic classifiers	scikit-learn

Time Series

Project	What it does	Stack
missing-value (time series)	Missing-value imputation for the monthly-sunspots series (Conv1D, GRU, BiGRU)	TensorFlow/Keras

Video Understanding

Project	What it does	Stack
hockey-videos	Hockey-fight detection with ResNet50 features + classifier	TensorFlow/Keras

Training Toolkit

Project	What it does	Stack
call back	Keras callbacks playbook: early stopping, checkpoints, LR scheduling on CIFAR-100 / Fashion-MNIST	TensorFlow/Keras
_BASICS	PyTorch fundamentals: tensors, CNN, RNN/LSTM, transfer learning, augmentation, custom datasets	PyTorch

Repository Layout

.
├── adversarial/                  # FGSM + DeepFool attacks
├── _BASICS/                      # PyTorch fundamentals
├── call back/                    # Keras callback patterns
├── FER/                          # Facial expression recognition
├── GAN/                          # GAN & DCGAN
├── hockey-videos/                # Fight detection in sports video
├── image_captioning/             # CNN-LSTM captioning
├── image-color-clustring/        # Color quantization
├── missing-value(time series)/   # Time-series imputation
├── _ML/sentiment_analysis_IMDB/  # IMDB sentiment analysis
├── pose estimation/              # Heatmap pose estimation
├── siameseNet/                   # Face recognition
├── STN-and-transferLearning/     # Spatial transformers + transfer learning
├── topic-modeling(NLP)/          # Persian news topic clustering
├── U-NET/                        # Binary segmentation
├── video-YOLO/                   # YOLOv5 driving-video analysis
└── vision-transformer/           # ViT / CCT from scratch

Tech Stack

• Languages: Python 3
• Deep learning: PyTorch, TensorFlow/Keras, timm
• Classical ML: scikit-learn, sentence-transformers
• Vision: OpenCV, albumentations
• Data & viz: NumPy, Pandas, Matplotlib, Seaborn
• Ops: Weights & Biases, TensorBoard, Jupyter

Getting Started

Each project is independent. To run one, install its dependencies and follow its README:

cd vision-transformer
pip install -r requirements.txt
python train.py --dataset cifar10 --model cct

Repository-wide environment

pip install torch torchvision tensorflow numpy pandas scikit-learn opencv-python \
            matplotlib seaborn einops timm sentence-transformers wandb

Several projects require external datasets (LFW, Flickr8k, CelebA, FER2013, LSP) — each README documents where to obtain them. Pretrained weights that ship in the repo (e.g. U-NET/binary_segmentation.h5, video-YOLO/yolov5s.pt) are used automatically.

Conventions

• Every project follows a consistent module layout: config.py, model.py, dataset.py, train.py, evaluate.py, utils.py, requirements.txt, and a README.md.
• All hyperparameters live in a single config.py.
• Training/evaluation run through argparse CLI entry points.
• The original .ipynb files are preserved unmodified as the research record.

License

MIT