model card: fix metrics (separate CFG 3 / 2.5 runs; T2M-GPT R@3 0.775)
Browse files
README.md
CHANGED
|
@@ -20,9 +20,13 @@ makes freezing exact, so the base model's language ability is fully preserved.
|
|
| 20 |
|
| 21 |
| Model | Params (trainable) | Frozen text LLM | R@1 | R@3 | FID |
|
| 22 |
|---|---|---|---|---|---|
|
| 23 |
-
| T2M-GPT | full | no | 0.492 | 0.
|
| 24 |
| MoMask | full | no | 0.521 | 0.807 | 0.045 |
|
| 25 |
-
| **ViBES-T2M (this)** | **0.27B** | **yes** | **0.
|
|
|
|
|
|
|
|
|
|
|
|
|
| 26 |
|
| 27 |
R@1 on par with T2M-GPT while keeping the entire conversational LLM intact.
|
| 28 |
|
|
|
|
| 20 |
|
| 21 |
| Model | Params (trainable) | Frozen text LLM | R@1 | R@3 | FID |
|
| 22 |
|---|---|---|---|---|---|
|
| 23 |
+
| T2M-GPT | full | no | 0.492 | 0.775 | 0.141 |
|
| 24 |
| MoMask | full | no | 0.521 | 0.807 | 0.045 |
|
| 25 |
+
| **ViBES-T2M (this), CFG 3** | **0.27B** | **yes** | **0.486** | **0.769** | **0.265** |
|
| 26 |
+
| ViBES-T2M (this), CFG 2.5 | 0.27B | yes | 0.489 | 0.766 | 0.301 |
|
| 27 |
+
|
| 28 |
+
Baseline rows are the numbers reported in their papers. ViBES-T2M rows sample with temperature 0.7 /
|
| 29 |
+
top-p 0.9 at the given classifier-free guidance scale; MM-Dist is 3.196 (CFG 3) and 3.208 (CFG 2.5).
|
| 30 |
|
| 31 |
R@1 on par with T2M-GPT while keeping the entire conversational LLM intact.
|
| 32 |
|