Upload README.md with huggingface_hub

README.md

@@ -23,16 +23,10 @@ ViT-g / 1.15B model), exposing a single-image **relative-depth** output for macO
   `int()` on size-1 non-0-dim arrays, which breaks the const-cast of `H//patch_size`).
   Single-image behaviour is unchanged.
 
-## Benchmark (vs the smaller CoreML depth models, 504×504, Apple Silicon, ALL compute units)
-
-| Model | Params | Latency | Depth corr. vs GIANT |
-|---|---|---|---|
-| DA3MONO-LARGE | 0.35B | ~0.32 s/img | 0.89–0.99 |
-| DA3-LARGE | 0.30B | ~0.34 s/img | 0.91–0.97 |
-| **DA3-GIANT** | **1.15B** | **~1.49 s/img** | — |
-
-GIANT is ~**4.4× slower** than the large/mono models for depth that correlates **0.89–0.99**
-with them — diminishing returns for monocular depth, but the highest-capacity option.
+This is the highest-capacity DA3 depth variant, and correspondingly the slowest at inference —
+for real-time monocular depth the smaller
+[DA3MONO-LARGE](https://huggingface.co/depth-anything/DA3MONO-LARGE) /
+[DA3-LARGE](https://huggingface.co/sdkv2/DA3-LARGE-CoreML) are usually the better trade-off.
 
 ## License & attribution