|
--- |
|
library_name: transformers |
|
tags: |
|
- generated_from_trainer |
|
model-index: |
|
- name: SkLIP-masked |
|
results: [] |
|
license: mit |
|
datasets: |
|
- joshuachou/SkinCAP |
|
language: |
|
- en |
|
base_model: |
|
- allenai/scibert_scivocab_uncased |
|
- openai/clip-vit-base-patch32 |
|
pipeline_tag: feature-extraction |
|
--- |
|
|
|
<!-- This model card has been generated automatically according to the information the Trainer had access to. You |
|
should probably proofread and complete it, then remove this comment. --> |
|
|
|
# SkLIPA |
|
|
|
SkLIPA (Skin CLIP Anonimised) is a hybrid CLIP model finetuned on the [SkinCAP](https://hf.rst.im/datasets/joshuachou/SkinCAP), a multi-modal dermatology dataset annotated |
|
with rich medical captions. It is built with a SciBERT text encoder and the pre-trained CLIP-32 vision encoder. |
|
|
|
The anonymisation procedure was designed to remove age and gender information form the textual description of each image, replacing them with `[AGE]` and `[GENDER]` tokens. |
|
|
|
## Model description |
|
|
|
More information needed |
|
|
|
## Intended uses & limitations |
|
|
|
More information needed |
|
|
|
## Training and evaluation data |
|
|
|
More information needed |
|
|
|
## Training procedure |
|
|
|
### Training hyperparameters |
|
|
|
The following hyperparameters were used during training: |
|
- learning_rate: 1e-05 |
|
- train_batch_size: 64 |
|
- eval_batch_size: 64 |
|
- seed: 42 |
|
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 |
|
- lr_scheduler_type: cosine |
|
- lr_scheduler_warmup_ratio: 0.01 |
|
- num_epochs: 10 |
|
|
|
### Training results |
|
|
|
| Training Loss | Epoch | Step | Validation Loss | |
|
|:-------------:|:-----:|:----:|:---------------:| |
|
| 4.2018 | 1.0 | 57 | 4.1344 | |
|
| 4.1697 | 2.0 | 114 | 4.1298 | |
|
| 4.1668 | 3.0 | 171 | 4.1276 | |
|
| 4.164 | 4.0 | 228 | 4.1263 | |
|
| 4.158 | 5.0 | 285 | 4.1253 | |
|
| 4.1583 | 6.0 | 342 | 4.1246 | |
|
| 4.1569 | 7.0 | 399 | 4.1243 | |
|
| 4.1575 | 8.0 | 456 | 4.1241 | |
|
| 4.1564 | 9.0 | 513 | 4.1240 | |
|
| 4.1604 | 10.0 | 570 | 4.1240 | |
|
|
|
|
|
### Framework versions |
|
|
|
- Transformers 4.45.2 |
|
- Pytorch 2.1.0+cu118 |
|
- Datasets 3.0.1 |
|
- Tokenizers 0.20.1 |
