Upload Archetecture.md
Browse files- DOCS/Archetecture.md +16 -0
DOCS/Archetecture.md
ADDED
|
@@ -0,0 +1,16 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
| **Component** | **Description** |
|
| 2 |
+
|-------------------------------|-----------------------------------------------------------------------------------------------|
|
| 3 |
+
| **Backbone** | ResNet-50 with FPN (Feature Pyramid Network) |
|
| 4 |
+
| **Pretrained Weights** | Trained on ImageNet for feature extraction. |
|
| 5 |
+
| **RPN (Region Proposal Network)** | Generates region proposals based on extracted features from the backbone. |
|
| 6 |
+
| **ROI Align** | Aligns region proposals to a fixed size for consistent feature extraction. |
|
| 7 |
+
| **Box Head** | Fully connected layers for refining bounding boxes and classifying objects. |
|
| 8 |
+
| **Box Predictor** | Replaced with a custom predictor: `FastRCNNPredictor` for handling custom classes. |
|
| 9 |
+
| **Number of Classes** | Configurable (including background). |
|
| 10 |
+
| **Loss Function** | Combines classification and regression losses for multi-task optimization. |
|
| 11 |
+
| **Optimizer** | Stochastic Gradient Descent (SGD) with momentum for optimization. |
|
| 12 |
+
| **Learning Rate Scheduler** | StepLR to decay learning rate every few epochs for better convergence. |
|
| 13 |
+
| **Batch Normalization** | Applied within the backbone for stable training. |
|
| 14 |
+
| **Data Format** | Input: Tensor of shape `(Batch Size, Channels, Height, Width)` in PyTorch's NCHW format. |
|
| 15 |
+
| **Output** | - Class probabilities for each region proposal. |
|
| 16 |
+
| | - Refined bounding box coordinates for each detected object. |
|