Closed BIGBALLON closed 9 months ago
Hello, I can list the results here if you want, but the classification performance of InternVL-G is not improved compared to InternVL-C. The main improvement of InternVL-G is retrieval capability.
Yes, I'm interested in how it performs on downstream Zeroshot classification tasks.
model name | IN-1K | IN-A | IN-R | IN-V2 | IN-Sketch | ObjectNet | ∆ | average |
---|---|---|---|---|---|---|---|---|
InternVL-C | 83.2 | 83.8 | 95.5 | 77.3 | 73.9 | 80.6 | 0.8 | 82.4 |
InternVL-G | - | - | - | - | - | - | - | - |
@BIGBALLON This is the result.
model name | IN-1K | IN-A | IN-R | IN-V2 | IN-Sketch | ObjectNet | ∆ | average |
---|---|---|---|---|---|---|---|---|
InternVL-C | 83.2 | 83.8 | 95.5 | 77.3 | 73.9 | 80.6 | 0.8 | 82.4 |
InternVL-G | 82.9 | 82.4 | 95.5 | 77.1 | 73.3 | 79.3 | 1.1 | 81.8 |
@czczup thanks !!!
I can list the results here if you want, but the classification performance of InternVL-G is not improved compared to InternVL-C. The main improvement of InternVL-G is retrieval capability.
Hi, @czczup, can you explain why InternVL-G improved retrieval capability?
Hi, Are there any Zeroshot classification (ImageNet) evaluation results for InternVL-G?