|
--- |
|
license: mit |
|
datasets: |
|
- imagenet-1k |
|
metrics: |
|
- accuracy |
|
pipeline_tag: image-classification |
|
--- |
|
# mrl-convnext-tiny |
|
|
|
We provide a ConvNeXt-Tiny model trained with [Matryoshka Representation Learning](https://github.com/RAIVNLab/MRL) (MRL) from scratch on ImageNet-1K, |
|
adapted from [the original repository](https://github.com/facebookresearch/ConvNeXt). We train with a Matryoshka Loss at [d] = {48, 96, 192, 384, 786}, |
|
|
|
<p align="center"> |
|
<img src="https://cdn-uploads.huggingface.co/production/uploads/64af72d4a609b29cc7b5919b/eGorc5hWdriNcEJkwP7Lz.png" alt="drawing" width="600"/> |
|
Top-1 Retrieval Accuracy (1-nearest neighbor) of mrl-convnext-tiny on ImageNet-1K with Exact search and IVF search across vector dimensionality. |
|
</p> |
|
|
