|
--- |
|
license: apache-2.0 |
|
language: |
|
- en |
|
library_name: open_clip |
|
pipeline_tag: zero-shot-image-classification |
|
--- |
|
|
|
# GenshinCLIP |
|
A simple open-sourced SigLIP model fine-tuned on Genshin Impact's image-text pairs. |
|
|
|
The model is far from being perfect, but could still offer some better text-image matching performance in some Genshin Impact scenarios. |
|
|
|
|
|
## Case Study |
|
|
|
| Case | Image | Multiple Choices | CLIPScore<br/>(After Finetune) | CLIPScore<br/>(Before Finetune) | |
|
|------|----------------------------------------------------------------------------------------------------------------------------------------------------|-----------------------------------------------------------------------------------------------------------------------------------------|--------------------------------------------------|---------------------------------------------------------------------| |
|
| 1 | <img src="https://static.wikia.nocookie.net/gensin-impact/images/2/24/Ganyu_Card.png/revision/latest?cb=20230519012433" height="64"> | 1) This is an image of Kuki Shinobu<br/>2) This is an image of Ganyu<br/>3) This is an image of Keqing<br/>4) This is an image of Liyue | 1) 0.000<br>2) **0.438**<br>3) 0.000<br>4) 0.000 | 1) **1.207e-06**<br/>2) 4.144e-08<br/>3) 1.201e-07<br/>4) 2.212e-08 | |
|
| 2 | <img src="https://static.wikia.nocookie.net/gensin-impact/images/3/33/Qingce_Village.png/revision/latest?cb=20220626161951" height="64"> | 1) This is an area of Liyue<br/>2) This is an area of Mondstadt<br/>3) This is an area of Sumeru<br/>4) This is Qingce Village | 1) 0.016<br>2) 0.000<br>3) 0.001<br>4) **0.233** | 1) 0.015<br/>2) 0.003<br/>3) 0.009<br/>4) **0.377** | |
|
| 3 | <img src="https://static.wikia.nocookie.net/gensin-impact/images/0/0c/Enemy_Boreas.png/revision/latest?cb=20210426192800" height="64"> | 1) This is Andrius Wolf of the North<br/>2) This is Stormterror Dvalin<br/>3) This is Guardian of Apep's Oasis | 1) 1.000<br>2) 0.000<br>3) 0.000 | 1) **0.001**<br/>2) 0.000<br/>3) 0.000 | |
|
| 4 | <img src="https://static.wikia.nocookie.net/gensin-impact/images/2/24/Item_Amakumo_Fruit_Wild.png/revision/latest?cb=20220601235905" height="64"> | 1) This is Amakumo Fruit<br/>2) This is Padisarahs<br/>3) This is Naku Weeds | 1) 0.000<br>2) 0.000<br>3) **0.024** | 1) 9.425e-07<br/>2) 1.207e-06<br/>3) **1.669e-05** | |
|
|
|
Note: Case 4 is a bad case. (Correct answer is Amakumo Fruit) |
|
|
|
|
|
## Intended uses & limitations |
|
|
|
You can use the raw model for tasks like zero-shot image classification and image-text retrieval. |
|
|
|
### How to use (With OpenCLIP) |
|
|
|
Here is how to use this model to perform zero-shot image classification: |
|
|
|
```python |
|
import torch |
|
import torch.nn.functional as F |
|
from PIL import Image |
|
import requests |
|
from open_clip import create_model_from_pretrained, get_tokenizer |
|
|
|
def preprocess_text(string): |
|
return "Genshin Impact\n" + string |
|
|
|
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu") |
|
|
|
# load checkpoint from local path |
|
# model_path = "path/to/open_clip_pytorch_model.bin" |
|
# model_name = "ViT-SO400M-14-SigLIP-384" |
|
# model, preprocess = create_model_from_pretrained(model_name=model_name, pretrained=model_path, device=device) |
|
# tokenizer = get_tokenizer(model_name) |
|
|
|
# or load from hub |
|
model, preprocess = create_model_from_pretrained('hf-hub:mrzjy/GenshinImpact-ViT-SO400M-14-SigLIP-384') |
|
tokenizer = get_tokenizer('hf-hub:mrzjy/GenshinImpact-ViT-SO400M-14-SigLIP-384') |
|
|
|
# image |
|
image_url = "https://static.wikia.nocookie.net/gensin-impact/images/3/33/Qingce_Village.png" |
|
image = Image.open(requests.get(image_url, stream=True).raw) |
|
image = preprocess(image).unsqueeze(0).to(device) |
|
|
|
# text choices |
|
labels = [ |
|
"This is an area of Liyue", |
|
"This is an area of Mondstadt", |
|
"This is an area of Sumeru", |
|
"This is Qingce Village" |
|
] |
|
labels = [preprocess_text(l) for l in labels] |
|
text = tokenizer(labels, context_length=model.context_length).to(device) |
|
with torch.autocast(device_type=device.type): |
|
with torch.no_grad(): |
|
image_features = model.encode_image(image) |
|
text_features = model.encode_text(text) |
|
image_features = F.normalize(image_features, dim=-1) |
|
image_features = F.normalize(image_features, dim=-1) |
|
text_features = F.normalize(text_features, dim=-1) |
|
text_probs = torch.sigmoid(image_features @ text_features.T * model.logit_scale.exp() + model.logit_bias) |
|
scores = [f"{s:.3f}" for i, s in enumerate(text_probs.tolist()[0])] |
|
print(scores) # [0.016, 0.000, 0.001, 0.233] |
|
``` |
|
|
|
## Model Card |
|
### SigLIP for GenshinImpact |
|
|
|
[SigLIP model](https://huggingface.co/timm/ViT-SO400M-14-SigLIP-384) further fine-tuned on 17k Genshin Impact English text-image pairs at resolution 384x384. |
|
|
|
### Training data description |
|
|
|
There're currently 17,428 (train) and 918 (validation) text-image pairs used for model training. |
|
|
|
All the images and texts are crawled from [Genshin Fandom Wiki](https://genshin-impact.fandom.com/wiki) and are manually parsed to form text-image pairs. |
|
|
|
**Image Processing:** |
|
- Size: Resize all images to 384x384 pixels to match the original model training settings. |
|
- Format: Accept images in PNG or GIF format. For GIFs, extract a random frame to create a static image for text-image pairs. |
|
|
|
**Text Processing:** |
|
- Source: Text can be from the simple caption attribute of an HTML `<img>` tag or specified web content. |
|
- Format: Prepend all texts with "Genshin Impact" along with some simple template to form natural language sentences. |
|
|
|
For example, here are some training image-text pairs: |
|
|
|
| Image | Text | |
|
|----------------------------------------------------------------------|-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------| |
|
| <img src="img/9361c6fe-dabe-43b3-ae76-9731e4b73e1b.png" height="64"> | Genshin Impact<br/>This is Tainted Water-Splitting Phantasm.<br/>Tainted Hydro Phantasm Idle Waving 2 | |
|
| <img src="img/911d0f58-dc35-4dfb-9afc-79a2d582881b.png" height="64"> | Genshin Impact<br/>This is Annihilation Specialist Mek.<br/>Although the Fontaine Research Institute of Kinetic Energy Engineering... ..., state apparatus enforcing the monopoly on violence. | |
|
| <img src="img/bd94b015-153f-41de-b8d4-049ba29481e3.png" height="64"> | Genshin Impact<br/>Collei<br/>Expressions<br/>a Smiley expression | |
|
| <img src="img/36a956e1-e6af-43bc-a71b-a893fe1bcf9e.png" height="64"> | Genshin Impact<br/>This is the map of viewpoint Nine Pillars of Peace in Cuijue Slope, Minlin, Liyue<br/>Nine shackles of stone were said to have been laid down deep in the valleys of Cuijue Slope to drive off evil and cleanse the world. | |
|
|
|
**Data Distribution:** |
|
|
|
 |
|
|
|
**Validation Loss Curve** |
|
|
|
 |
|
|
