Update README.md

README.md

@@ -28,9 +28,7 @@ This repository contains lightweight vision-centric Bird’s‐Eye View (BEV) pe
 
 ## Model Architectures
 
-RegNetX Backbone → FPN Neck → 6× Transformer Decoder Layers
-                                 ↓
-               BEV Queries (learned or position-aware) → Detection & Segmentation Heads
+RegNetX Backbone → FPN Neck → 6× Transformer Decoder Layers → BEV Queries (learned or position-aware) → Detection & Segmentation Heads
 
 - **Depth‐Aware Fusion (DETR3D)**: Implicitly learns spatial alignment across views via cross‐attention.
 - **Position-Aware Embeddings (PETR)**: Encodes camera geometry for enhanced BEV query initialization.