Model Card for tmp-pl-image-classification

์ด ์ €์žฅ์†Œ๋Š” ๐Ÿค— Transformers์˜ pipeline() ๋™์ž‘์„ ์ดํ•ดํ•˜๊ณ  ์—ฐ์Šตํ•˜๊ธฐ ์œ„ํ•œ ํ•™์Šต์šฉ(pipeline practice) ๋ชจ๋ธ repo ์ž…๋‹ˆ๋‹ค.
๋ชจ๋ธ ๊ฐ€์ค‘์น˜๋Š” ์›๋ณธ ๋ชจ๋ธ google/vit-base-patch16-224 ์„ ๊ทธ๋Œ€๋กœ ์‚ฌ์šฉํ•˜๋ฉฐ, ์ถ”๊ฐ€์ ์ธ fine-tuning์€ ์ˆ˜ํ–‰ํ•˜์ง€ ์•Š์•˜์Šต๋‹ˆ๋‹ค.


Model Details

Model Description

๋ณธ ๋ชจ๋ธ์€ Vision Transformer(ViT) ๊ธฐ๋ฐ˜ ์ด๋ฏธ์ง€ ๋ถ„๋ฅ˜ ๋ชจ๋ธ์„ pipeline("image-classification") ํ˜•ํƒœ๋กœ
Hub์— ์—…๋กœ๋“œํ•˜๊ณ  ๋‹ค์‹œ ๋ถˆ๋Ÿฌ์˜ค๋Š” ์ „์ฒด ํ๋ฆ„์„ ์‹ค์Šตํ•˜๊ธฐ ์œ„ํ•ด ๊ตฌ์„ฑ๋จ.

  • Developed by: Google Research (์›๋ณธ๋ชจ๋ธ)
  • Shared by [optional]: dsaint31
  • Model type: Image Classification (Vision Transformer)
  • Language(s) (NLP): ํ•ด๋‹น ์—†์Œ (์ด๋ฏธ์ง€ ์ž…๋ ฅ)
  • License: Apache-2.0
  • Finetuned from model [optional]: google/vit-base-patch16-224 (๊ฐ€์ค‘์น˜ ๋ณ€๊ฒฝ ์—†์Œ. fine-tuning ๋ฏธ์ˆ˜ํ–‰)

Model Sources [optional]

Uses

Direct Use

  • pipeline("image-classification", model=...) ์‚ฌ์šฉ๋ฒ• ์‹ค์Šต
  • Hugging Face Hub์— pipeline ํ˜•ํƒœ๋กœ ๋ชจ๋ธ์„ ์—…๋กœ๋“œ / ๋‹ค์šด๋กœ๋“œํ•˜๋Š” ํ๋ฆ„ ์ดํ•ด
  • Vision ๋ชจ๋ธ๊ณผ pipeline์˜ ๊ด€๊ณ„ ํ•™์Šต

Downstream Use [optional]

  • ๋ณธ repo ์ž์ฒด๋Š” downstream task๋ฅผ ์œ„ํ•œ fine-tuning์„ ๋ชฉ์ ์œผ๋กœ ํ•˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค.
  • ํ•™์Šต ๋˜๋Š” ์„ฑ๋Šฅ ๋น„๊ต ๋ชฉ์ ์ด๋ผ๋ฉด ์›๋ณธ ๋ชจ๋ธ repo๋ฅผ ์ง์ ‘ ์‚ฌ์šฉํ•˜๋Š” ๊ฒƒ์ด ์ ์ ˆํ•ฉ๋‹ˆ๋‹ค.

Out-of-Scope Use

  • ๋ชจ๋ธ ์„ฑ๋Šฅ ํ‰๊ฐ€ ๋˜๋Š” ๋ฒค์น˜๋งˆํฌ
  • ์‹ค์ œ ์„œ๋น„์Šค ํ™˜๊ฒฝ์—์„œ์˜ ๋ชจ๋ธ ๋ฐฐํฌ
  • ํŠน์ • ๋„๋ฉ”์ธ(์˜๋ฃŒ, ์‚ฐ์—… ์˜์ƒ ๋“ฑ)์— ๋Œ€ํ•œ ์‹ ๋ขฐ์„ฑ ์žˆ๋Š” ์ถ”๋ก 

Bias, Risks, and Limitations

  • ๋ณธ ๋ชจ๋ธ์€ ImageNet ๊ธฐ๋ฐ˜ ๋ฐ์ดํ„ฐ๋กœ ํ•™์Šต๋œ ์ผ๋ฐ˜ ๋ชฉ์  ์ด๋ฏธ์ง€ ๋ถ„๋ฅ˜ ๋ชจ๋ธ์˜ ํŠน์„ฑ์„ ๊ทธ๋Œ€๋กœ ๊ฐ€์ง‘๋‹ˆ๋‹ค.
  • ํŠน์ • ๊ฐ์ฒด, ๋ฌธํ™”์  ๋งฅ๋ฝ, ์ „๋ฌธ ๋„๋ฉ”์ธ์— ๋Œ€ํ•œ ๋ถ„๋ฅ˜ ์„ฑ๋Šฅ์€ ๋ณด์žฅ๋˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค.
  • ๋ณธ repo๋Š” ์—ฐ์Šต์šฉ pipeline ์ €์žฅ์†Œ์ด๋ฏ€๋กœ ๋ชจ๋ธ์˜ ์‚ฌํšŒ์  ์˜ํ–ฅ์ด๋‚˜ ํŽธํ–ฅ ๋ถ„์„์„ ๋ชฉ์ ์œผ๋กœ ํ•˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค.

Recommendations

Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.

  • ์‹ค์ œ ์‚ฌ์šฉ ๋ชฉ์ ์ด ์žˆ๋Š” ๊ฒฝ์šฐ, ์›๋ณธ ๋ชจ๋ธ ์นด๋“œ(google/vit-base-patch16-224)์˜ ์ œํ•œ ์‚ฌํ•ญ์„ ๋ฐ˜๋“œ์‹œ ์ฐธ๊ณ ํ•˜์‹ญ์‹œ์˜ค.
  • ์ด repo๋Š” ํ•™์Šต ๋ฐ ์‹ค์Šต ๋ชฉ์ ์— ํ•œํ•ด ์‚ฌ์šฉํ•˜๊ธฐ๋ฅผ ๊ถŒ์žฅํ•ฉ๋‹ˆ๋‹ค.

How to Get Started with the Model

Use the code below to get started with the model.

์•„๋ž˜ ์˜ˆ์ œ๋Š” Hugging Face pipeline์„ ์ด์šฉํ•ด ๋ณธ ๋ชจ๋ธ์„ ๋กœ๋“œํ•˜๊ณ  ์ด๋ฏธ์ง€๋ฅผ ๋ถ„๋ฅ˜ํ•˜๋Š” ์ตœ์†Œ ์˜ˆ์ œ์ž…๋‹ˆ๋‹ค.

from transformers import pipeline
from PIL import Image
import requests

img_url = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/cats.png"
image = Image.open(requests.get(img_url, stream=True).raw)

clf = pipeline(
    task="image-classification",
    model="dsaint31/tmp-pl-image-classification",
)

print(clf(image))

[More Information Needed]

Training Details

Training Data

  • ๋ณธ repo์—์„œ๋Š” ์ถ”๊ฐ€ ํ•™์Šต์„ ์ˆ˜ํ–‰ํ•˜์ง€ ์•Š์•˜์Šต๋‹ˆ๋‹ค.
  • ์›๋ณธ ๋ชจ๋ธ์€ ImageNet-21k๋กœ ์‚ฌ์ „ํ•™์Šต(pretraining) ํ›„ ImageNet-1k๋กœ fine-tuning๋œ ๋ชจ๋ธ์ž…๋‹ˆ๋‹ค.

Training Procedure

Preprocessing [optional]

  • ์›๋ณธ ViT ๋ชจ๋ธ์˜ ๊ธฐ๋ณธ ์ด๋ฏธ์ง€ ์ „์ฒ˜๋ฆฌ(Image Processor)๋ฅผ ๊ทธ๋Œ€๋กœ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค.

[More Information Needed]

Training Hyperparameters

  • Training regime: ํ•ด๋‹น ์—†์Œ(ํ•™์Šต ๋ฏธ์ˆ˜ํ–‰) [More Information Needed]

Speeds, Sizes, Times [optional]

Evaluation

Testing Data, Factors & Metrics

Testing Data

  • ๋ณธ repo์—์„œ๋Š” ๋ณ„๋„์˜ ํ‰๊ฐ€๋ฅผ ์ˆ˜ํ–‰ํ•˜์ง€ ์•Š์•˜์Šต๋‹ˆ๋‹ค.
  • ์„ฑ๋Šฅ ์ง€ํ‘œ๋Š” ์›๋ณธ ๋ชจ๋ธ ์นด๋“œ์˜ ํ‰๊ฐ€ ๊ฒฐ๊ณผ๋ฅผ ์ฐธ๊ณ ํ•˜์‹ญ์‹œ์˜ค.

[More Information Needed]

Factors

[More Information Needed]

Metrics

[More Information Needed]

Results

[More Information Needed]

Summary

Model Examination [optional]

[More Information Needed]

Environmental Impact

๋ณธ repo์—์„œ๋Š” ํ•™์Šต์„ ์ˆ˜ํ–‰ํ•˜์ง€ ์•Š์•˜์œผ๋ฏ€๋กœ ์ถ”๊ฐ€์ ์ธ ํ™˜๊ฒฝ์  ์˜ํ–ฅ์€ ์—†์Šต๋‹ˆ๋‹ค.

  • ์›๋ณธ ๋ชจ๋ธ ํ•™์Šต์— ๋Œ€ํ•œ ํ™˜๊ฒฝ ์˜ํ–ฅ์€ base model ๋ฌธ์„œ๋ฅผ ์ฐธ๊ณ ํ•˜์‹ญ์‹œ์˜ค.

Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).

  • Hardware Type: [More Information Needed]
  • Hours used: [More Information Needed]
  • Cloud Provider: [More Information Needed]
  • Compute Region: [More Information Needed]
  • Carbon Emitted: [More Information Needed]

Technical Specifications [optional]

Model Architecture and Objective

  • Vision Transformer (ViT-Base, patch size 16, input resolution 224x224)
  • Objective: Image classification

Compute Infrastructure

[More Information Needed]

Hardware

  • ํ•ด๋‹น์—†์Œ (ํ•™์Šต ๋ฏธ์ˆ˜ํ–‰)

Software

  • Transformers
  • Pillow
  • PyTorch

[More Information Needed]

Citation [optional]

์›๋ณธ ๋ชจ๋ธ ์ธ์šฉ ์‹œ ์•„๋ž˜ ๋…ผ๋ฌธ์„ ์ฐธ๊ณ ํ•˜์‹ญ์‹œ์˜ค.

BibTeX:

@article{dosovitskiy2020image,
  title={An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale},
  author={Dosovitskiy, Alexey and others},
  journal={arXiv preprint arXiv:2010.11929},
  year={2020}
}

APA:

[More Information Needed]

Glossary [optional]

[More Information Needed]

More Information [optional]

[More Information Needed]

Model Card Authors [optional]

  • dsaint31 (pipeline practice repository)

Model Card Contact

Downloads last month
28
Safetensors
Model size
86.6M params
Tensor type
F32
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for dsaint31/tmp-pl-image-classification

Finetuned
(1944)
this model

Dataset used to train dsaint31/tmp-pl-image-classification

Papers for dsaint31/tmp-pl-image-classification