Update README.md

README.md

@@ -1,3 +1,31 @@
----
-license: mit
----
+---
+license: mit
+tags:
+- implicit-neural-representation
+- siren
+- image-compression
+- coordinates-based-model
+- computer-vision
+model_type: SIREN
+---
+
+# Neural RAM Representation (SIREN)
+
+This model is an **Implicit Neural Representation (INR)** of a RAM module image. Unlike traditional raster formats (PNG/JPG) that store pixels in a grid, this image is encoded entirely within the weights of a neural network.
+
+## Project Overview
+The model utilizes the **SIREN** (Sinusoidal Representation Networks) architecture. It is trained to map continuous 2D coordinates $(x, y)$ to their corresponding $RGB$ color values. 
+
+### Key Features:
+- **Mathematical Storage:** The image exists as a continuous function rather than a discrete pixel array.
+- **Infinite Resolution:** You can render the image at any scale by increasing the density of the coordinate grid.
+- **Efficient Encoding:** The model has been optimized (~12,000 epochs) to capture fine textures of the PCB and gold-plated contacts with high fidelity.
+
+## Technical Specifications
+- **Architecture:** Multi-Layer Perceptron (MLP) with periodic (sine) activation functions.
+- **Depth:** 2 hidden layers.
+- **Width:** 128 neurons per layer.
+- **Format:** Provided as a lightweight `.safetensors` file.
+
+## Usage
+To render the image, you will need `torch`, `safetensors`, and `numpy`.