image search using CLIP/Chinese-CLIP feature.
pip3 install -r requirements.txt
modify the IMAGE_PATH in config.py and run
python3 extract_images_features.py
python3 app.py
- image feature: 768
- image enter
- filename, extension, height, width, filesize, createtime, clipfeature
- memory-3.5kB/per(clip/cnclip feature)
- time-2.7w/1h
- image-image threshold: 0.6
- text-image threshold: p-0.22/n-0.18
- cn text-image threshold: p-0.30/n-0.30
- match: cosine
- extract images feature(image enter)
- prompt(positive/negative)/image search
- positive/negative word-segmentation: use ";" to split prompt(e.g. yellow; person;)
- chinese
- OCR/chinese-OCR: english-"<text>artificial intelligence"; chinese-"<文本>人工智能"
- match/ocr-match algorithm