4 ms·
It is exciting that you could train a CLIP-style model from scratch with only 4M datapoints. But if you’ve got that data, why not fine tune a pretrained model w
by juxtaposicion 4y ago
It is exciting that you could train a CLIP-style model from scratch with only 4M datapoints. But if you’ve got that data, why not fine tune a pretrained model with your 4M points? It seems likely to outperform the from-scratch method.
- vov_or 4y agoThere is not only a difference in the data source but pre-trained tasks as well. But you are right, a fine-tuned models on human-annotated data are way better than zero-shot (just pre-trained) on Image retrieval. And it is correct for CLIP, ALBEF, VICHA, and UFORM.
- riku_iki 4y agoperhaps this approach can lead to better training of foundational models?..
- vov_or 4y agoMore efficient - for sure!