Hugging Face's logo

luisf-mc
/
gui-g2-3b-ccf

Image-to-Text
Transformers
English
gui-grounding
screen-understanding
vision-language-model
icon-detection
screenspot
visual-search
Model card Files
xet
 Community
luisf-mc commited on
Commit
4ed6594
·
verified ·
1 Parent(s): 601ce21

Update live demo perf: 2s fast mode, 8s accurate

Browse files
Files changed (1) hide show
  1. README.md +1 -1
README.md CHANGED
@@ -23,7 +23,7 @@ metrics:
23
  A drop-in inference wrapper that improves GUI-G2-3B's icon-grounding accuracy by **+2.2pp on ScreenSpot-v2** at zero training cost. The base weights are unchanged; everything is in the inference pipeline.
24
 
25
  > **Try it live (Azure A100, scale-to-zero)**: <https://guigrounding.whiteplant-27564a0e.eastus.azurecontainerapps.io>
26
- > Cold start ~90s the first time after idle, then ~6s/request (fast mode) or ~30s/request (accurate mode with 6-pass self-consistency).
27
 
28
  ![Side-by-side: 4 ScreenSpot-v2 icons where GUI-G2-3B baseline misses (red X) and CCF hits (green check). Blue boxes are ground truth.](comparison.png)
29
 
 
23
  A drop-in inference wrapper that improves GUI-G2-3B's icon-grounding accuracy by **+2.2pp on ScreenSpot-v2** at zero training cost. The base weights are unchanged; everything is in the inference pipeline.
24
 
25
  > **Try it live (Azure A100, scale-to-zero)**: <https://guigrounding.whiteplant-27564a0e.eastus.azurecontainerapps.io>
26
+ > Warm latency: **~2s/request (fast mode)** with CCF refinement, or ~8s for accurate mode (7-pass self-consistency, real agreement-based confidence). Cold start ~90s the first time after idle.
27
 
28
  ![Side-by-side: 4 ScreenSpot-v2 icons where GUI-G2-3B baseline misses (red X) and CCF hits (green check). Blue boxes are ground truth.](comparison.png)
29