en_ner_prompting / README.md
teo-sanchez's picture
Update README.md
43ea6d3
metadata
tags:
  - spacy
  - token-classification
language:
  - en
model-index:
  - name: en_ner_prompting
    results:
      - task:
          name: NER
          type: token-classification
        metrics:
          - name: NER Precision
            type: precision
            value: 0.7437641723
          - name: NER Recall
            type: recall
            value: 0.7248618785
          - name: NER F Score
            type: f_score
            value: 0.7341913822
widget:
  - text: >-
      Golden statue of a victorious warrior raising his sword to the sky,
      heroic, glorious, in the style of artgerm, gerald brom, atey ghailan and
      mike mignola, vibrant colors and hard shadows and strong rim light, plain
      background, comic cover art, trending on artstation
  - text: >-
      Italian renaissance dragon statue castle gallery highly detailed
      artstation concept art sharp focus illustration briclot rutkowski mucha
  - text: >-
      Quetzalcoatl in an epic battle with garuda, fantasy, stained glass, d & d,
      intricate, elegant, highly detailed, digital painting, artstation, concept
      art, matte, sharp focus, illustration, art by john collier and albert
      aublet and krenz cushart and artem demura and alphonse mucha
lisence: CC BY 3.0
Feature Description
Name en_ner_prompting
Version 0.0.3
spaCy >=3.4.3,<3.5.0
Default Pipeline tok2vec, ner
Components tok2vec, ner
Vectors 514157 keys, 514157 unique vectors (300 dimensions)
Sources n/a
License CC BY 3.0
Author Selas.ai

Description

Name entity recognition model to analyzing text-to-image prompts (Stable Diffusion).

The entities comprise 7 main categories and 11 subcategories for a total of 16 categories, extracted from a topic analysis made with BERTopic. The topic analysis can be explored the following visualization.

  β”œβ”€β”€ medium/
  β”‚   β”œβ”€β”€ photography
  β”‚   β”œβ”€β”€ painting
  β”‚   β”œβ”€β”€ rendering
  β”‚   └── illustration
  β”œβ”€β”€ influence/
  β”‚   β”œβ”€β”€ artist
  β”‚   β”œβ”€β”€ genre
  β”‚   β”œβ”€β”€ artwork
  β”‚   └── repository
  β”œβ”€β”€ light
  β”œβ”€β”€ color
  β”œβ”€β”€ composition
  β”œβ”€β”€ detail
  └── context/
      β”œβ”€β”€ era
      β”œβ”€β”€ weather
      └── emotion

Prompt data are from the diffusionDB database and were annotated by hand using Prodigy.

Label Scheme

View label scheme (16 labels for 1 components)
Component Labels
ner color, composition, context/emotion, context/era, context/weather, detail, influence/artist, influence/artwork, influence/genre, influence/repository, light, medium/illustration, medium/painting, medium/photography, medium/rendering, subject

Accuracy

Type Score
ENTS_F 73.42
ENTS_P 74.38
ENTS_R 72.49
TOK2VEC_LOSS 19323.84
NER_LOSS 144524.82