AdamAI-Systems — Vision

Building computer vision systems that perceive, analyze, and understand the visual world. This repository houses the image-classification, object-detection, segmentation, and visual-recognition solutions developed by AdamAI-Systems.

Projects in this Repository

1. Archaeological Landmark Classifier

A deep learning classifier that identifies archaeological landmarks from images.

• Tech Stack: Python, PyTorch (VGG19 / ResNet50), Streamlit.
• Key Features: Transfer learning on multiple backbones, interactive Streamlit web app, CLI prediction script.

2. Face Age & Gender Estimation

A multi-task CNN that simultaneously predicts age (regression) and gender (binary classification) from facial images.

• Tech Stack: Python, TensorFlow / Keras, UTKFace dataset.
• Key Features: Joint age + gender heads, end-to-end Colab training pipeline.

3. Face Orientation Classification

A CNN-based classifier for detecting head orientation in five classes (Front / Left / Right / Up / Down).

• Tech Stack: Python, TensorFlow / Keras.
• Key Features: Five-way orientation classification, interactive upload and webcam demos.

4. Fire Detection (YOLOv8)

Real-time fire detection in images using state-of-the-art object detection.

• Tech Stack: Python, Ultralytics YOLOv8, KaggleHub dataset loader.
• Key Features: YOLOv8-based detection, training notebook with checkpoint/backup support, inference notebook.

5. Fruit & Vegetable Image Classifier

Custom Inception-style CNN classifying 36 fruit and vegetable categories.

• Tech Stack: Python, TensorFlow / Keras.
• Key Features: Inception-inspired architecture, 36-class fine-grained classification on the Kaggle fruit-and-vegetable dataset.

6. Indian Sign Language Recognition (35 classes)

A CNN-based recognizer for Indian Sign Language covering digits 1–9 and letters A–Z.

• Tech Stack: Python, TensorFlow / Keras, OpenCV.
• Key Features: Edge-enhancement preprocessing, 35-class classification, Colab-ready notebook.

7. Mask Detection (Transfer Learning)

Face-mask detection (with_mask / without_mask) using transfer learning.

• Tech Stack: Python, TensorFlow / Keras (MobileNetV2), Weights & Biases.
• Key Features: Transfer learning + fine-tuning, experiment tracking via W&B.

8. Mini Photoshop (OpenCV + Ultralytics)

A desktop "Mini Photoshop" with basic image editing operations and AI-based segmentation.

• Tech Stack: Python, Tkinter, OpenCV, Pillow, Ultralytics YOLOv8-Seg.
• Key Features: Crop / blur / rotate / flip / edges / brightness, K-Means segmentation, YOLOv8 segmentation.

9. Plant Leaf Health Classification

A PyTorch CNN classifying healthy plant leaves across 12 crop classes.

• Tech Stack: Python, PyTorch, Weights & Biases.
• Key Features: Custom LeafCNN, weighted CrossEntropy for class imbalance, W&B experiment logging.

10. Sign Language Recognition (PyTorch + Gradio)

American Sign Language alphabet recognition (29 classes) with a web demo.

• Tech Stack: Python, PyTorch (ResNet18 transfer learning), Gradio.
• Key Features: Transfer learning, Gradio interface supporting both image upload and webcam input.

11. Traffic Sign AI Demo

German traffic sign classification (GTSRB, 43 classes) with an interactive demo.

• Tech Stack: Python, TensorFlow / Keras, Streamlit, gTTS.
• Key Features: CNN classifier, Grad-CAM visualization, Arabic class labels, voice alerts via gTTS.

---

Setup & General Instructions

To run any of the projects locally, navigate to the specific project directory and follow the instructions in its respective README.md.

Most projects follow a common workflow:

1. Create a virtual environment (python -m venv .venv) and activate it.
2. Install the project's dependencies (pip install -r requirements.txt where available).
3. Open the provided notebook (*.ipynb) or run the entry-point script (e.g. app.py, main.py, predict_*.py).

Note: Model weights (.pt, .pth, .h5, .keras) and datasets are intentionally not committed to this repository. They are either downloaded automatically by their respective frameworks (e.g. Ultralytics YOLOv8) or must be obtained from the dataset source linked in each project's README.

---

© 2026 AdamAI-Systems. All rights reserved.