Refresh Space overview copy

README.md

@@ -12,7 +12,12 @@ short_description: 'Exploring spectrograms, LWM embedding, and its evaluations.'
 
 # LWM-Spectro Lab
 
-One-stop lab for exploring spectrograms, LWM embeddings, and lightweight evaluation baselines.
+Walk through the Sub-6 GHz → mmWave workflow end-to-end: pretrained LWM encoders digest Sub-6 GHz spectrograms, lightweight heads approximate the residual 1D-CNN beam scorer (~500k params), and MoE embeddings expose the joint SNR/mobility cues used to rank mmWave beams.
+
+**What you get**
+- **Scenario coverage:** Six DeepMIMO deployments (excluded from LWM pretraining) across LTE / WiFi / 5G with modulation, SNR, and mobility metadata for filtering.
+- **Pipeline checkpoints:** Inspect raw spectrograms, compare LWM embeddings vs. raw vectors via balanced t-SNE, and plug embeddings into k-NN prototypes before training the production beam-selection head.
+- **Evaluation hooks:** Joint SNR/Doppler MoE embeddings approximate the feature bank that conditions the mmWave codebook selector, enabling rapid benchmarking of routing strategies.
 
 ## Features
 - Visualize LWM embeddings or raw spectrograms with customizable filters.

app.py

@@ -745,17 +745,14 @@ with gr.Blocks(title="LWM-Spectro Lab") as demo:
     gr.Markdown("# 🔬 LWM-Spectro Interactive Demo")
     gr.Markdown(
         """
-    Compare **LWM embeddings** vs **Raw Spectrograms** for visualization, then evaluate **precomputed MoE embeddings**
-    with a lightweight k-NN prototype classifier for joint SNR/Doppler recognition.
-    """
-    )
-    gr.Markdown(
-        """
-    📡 **Sub-6 GHz → mmWave Beam Prediction Task**
-    - LWM extracts signal-centric features from Sub-6 GHz spectrograms, mirroring the first stage of the beam prediction pipeline.
-    - A lightweight downstream model (residual 1D-CNN, ≈500k params) consumes those embeddings to rank mmWave beams from the codebook.
-    - The bundled dataset mixes six DeepMIMO scenarios that were held out from LWM pre-training, highlighting cross-scenario generalization.
-    - Use the tabs below to inspect raw inputs, visualize the embedding space, and benchmark simple classifiers before deploying larger heads.
+    Walk through the Sub-6 GHz → mmWave workflow end-to-end: pretrained **LWM encoders** digest Sub-6 GHz spectrograms,
+    lightweight heads act as stand-ins for the residual 1D-CNN beam scorer (~500k params), and **MoE embeddings** expose
+    the joint SNR/mobility cues used to rank mmWave beams.
+
+    **What you get**
+    - **Scenario coverage:** Six DeepMIMO deployments (held out from LWM pretraining) spanning LTE / WiFi / 5G with modulation, SNR, and mobility metadata for filtering.
+    - **Pipeline checkpoints:** Inspect raw spectrograms, compare LWM embeddings vs. raw vectors via balanced t-SNE, and plug embeddings into k-NN prototypes before training the production beam-selection head.
+    - **Evaluation hooks:** Joint SNR/Doppler MoE embeddings approximate the feature bank that conditions the mmWave codebook selector, enabling rapid benchmarking of routing strategies.
     """
     )