Added Smooth-GmP - best fine-tune yet! ✨
Browse files
README.md
CHANGED
|
@@ -3,6 +3,22 @@ license: mit
|
|
| 3 |
datasets:
|
| 4 |
- SPRIGHT-T2I/spright_coco
|
| 5 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 6 |
## A fine-tune of OpenAI / CLIP ViT-L/14 that has an unprecedented ImageNet/ObjectNet accuracy of ~0.90 (original pre-trained model / OpenAI's CLIP: ~0.85)**.
|
| 7 |
|
| 8 |
Made possible with Geometric Parametrization (GmP):
|
|
|
|
| 3 |
datasets:
|
| 4 |
- SPRIGHT-T2I/spright_coco
|
| 5 |
---
|
| 6 |
+
## Update 11/AUG/2024:
|
| 7 |
+
|
| 8 |
+
New Best-Performing CLIP ViT-L/14 'GmP-smooth' model added (simply download the files named *BEST*!):
|
| 9 |
+
|
| 10 |
+
|
| 11 |
+
|
| 12 |
+
Or just create a fine-tune yourself: [https://github.com/zer0int/CLIP-fine-tune](https://github.com/zer0int/CLIP-fine-tune)
|
| 13 |
+
|
| 14 |
+
How?
|
| 15 |
+
- Geometric Parametrization (GmP) (same as before)
|
| 16 |
+
- Activation Value manipulation for 'adverb neuron' (same as before)
|
| 17 |
+
- NEW: Custom loss function with label smoothing!
|
| 18 |
+
- For in-depth details, see my GitHub. 🤗
|
| 19 |
+
|
| 20 |
+
----
|
| 21 |
+
|
| 22 |
## A fine-tune of OpenAI / CLIP ViT-L/14 that has an unprecedented ImageNet/ObjectNet accuracy of ~0.90 (original pre-trained model / OpenAI's CLIP: ~0.85)**.
|
| 23 |
|
| 24 |
Made possible with Geometric Parametrization (GmP):
|