🚀 YOLO & SSD: Complete Mathematical & Conceptual Guide 🟢 1. YOLO (You Only Look Once) — Core Philosophy Unlike R-CNN family (two-stage detection), YOLO is: 🎯 Single-stage, fully conv...

🚀 Object Detection Evolution: From R-CNN to Faster R-CNN 🧭 1. Object Detection: Formal Problem Definition Given an image: [I \in \mathbb{R}^{H \times W \times 3}] Goal: predict a variable-len...

Video Transformers This note covers three important families of transformer-based video models: 1) ViViT (Video Vision Transformer) — Model 1 to Model 4 2) TimeSFormer (Time-Space Transformer) 3)...

Vision Transformer (ViT) Patch Embedding Given image: [X \in \mathbb{R}^{H \times W \times 3}] Patch size: [P \times P] Number of patches: [N = \frac{HW}{P^2}] Flatten each patch: [x_i \in...

Video Representation & Fusion (Detailed Technical Notes) This document explains in detail: Multi-frame modeling Early fusion Late fusion Temporal pooling Optical flow (with full de...

📊 Data Preprocessing & Batch Normalization 📌 1. Data Preprocessing 🎯 Goal One major goal of preprocessing: Zero-centered data Unit-variance data For input $x$: [\tilde{x} = \frac{...

🚀 Optimization in Deep Learning From SGD → Momentum → AdaGrad → RMSProp → Adam → Second-Order 🎯 Why Optimization Matters We want to minimize: [J(\theta)] by iteratively updating parameters: ...

🧠 Regularization & Dropout (GitHub Safe Version) 1️⃣ Training Pipeline (Big Picture) Machine Learning is data-driven: Design model (e.g., neural network) Initialize parameters $W$ F...

🚀 Learning Rate & Scheduling — Complete Notes 1️⃣ How to Choose a Learning Rate All optimizers (SGD, Momentum, AdaGrad, RMSProp, Adam) require a learning rate $\alpha$ as a hyperparameter. ...

🔥 Weight Initialization in Deep Neural Networks 1️⃣ Small Gaussian Random Initialization Initialize weights as: [W_{ij} \sim \mathcal{N}(0, \sigma^2)] with [\sigma = 0.01] In practice: W =...