Update ReadMe

Signed-off-by: Benedikt Blumenstiel <[email protected]>

README.md

@@ -35,12 +35,18 @@ During pre-training, TerraMind leverages masked token reconstruction, learning c
 
 ## Evaluation
 
-![terramind_evaluation.png](assets%2Fterramind_evaluation.png)
+![terramind_tiny_radar.png](assets/terramind_tiny_radar.png)
 
 We benchmarked TerraMind against other geospatial foundation models using the PANGAEA benchmark. 
-TerraMind consistently achieved state-of-the-art performance, surpassing existing models in various downstream tasks such as land use segmentation, water body mapping, and vegetation assessments.
+TerraMind base and large consistently achieved state-of-the-art performance, surpassing existing models in various downstream tasks such as land use segmentation, water body mapping, and vegetation assessments.
 The evaluation highlights its effectiveness in handling diverse Earth Observation scenarios.
-We present additional experiments in our [pre-print](https://arxiv.org/abs/2504.11171).
+We present additional experiments in our [paper](https://arxiv.org/abs/2504.11171).
+
+![GFLOPs.png](assets/GFLOPs.png)
+
+While TerraMind (TM) base and large versions provide SOTA performance, our tiny and small version perform competitive with much smaller and faster backbones. 
+The diagram above compares the computational requirements (GFLOPs), model parameters (marker size), and downstream task performance.
+Tiny can run easily on edge devices such as satellites or smartphones.
 
 
 ## Usage