Description Generator Concept; CV Concept
- [Complete] Image Data input (jpg, etc.…)
- [Complete] Object detection: Fine-tune VGG166 Image Classification (subject, verb and confidence value)
- Build
- Train
- Predict
- [Complete] Scene detection: Keras + VGG16 + Places365 (place and confidence value)
- [incomplete] Category Filter
- [Complete] Fasttext & Textgrocery (Text Category and confidence value)
- [Progressing] LCS Confidence + Subject Confidence
- [incomplete] NLP - Generate Sentence & Probabilistic
- n-gram: [S V O P]
- NLTK
Recognization Data Format:
"Description": {
"tags": [{
"obj": "string",
"label": "string",
"confidence": value
}, …],
"caption": [{
"text": "string",
"confidence": value
}],
"category": [{
"name": "string",
"score": value
}]
}
Keras + CRFasRNN
- MaskRNN(attempt)...[Failed]
- subject + facial + action + scene
n-gram + NLTK
- Raw text Processing
- Sentence Segmentation (lists of strings)
- Tokenization (sentences)
- Categorizing and Tagging words (tokenized sentences)
- Extracting, recognize the entities (pos-tagged sentences)
- Analyzing sentence structure (chunked sentences)
- Build grammars (relations)
Result:
Subject & score
+
Verb & score
+
Object & score
+
Context & score
+
Place & score
Reference
- https://keras.io
- https://github.com/matterport/Mask_RCNN
- https://github.com/sadeepj/crfasrnn_keras
- https://github.com/oswaldoludwig/Human-Action-Recognition-with-Keras
- https://github.com/jalajthanaki/Facial_emotion_recognition_using_Keras
- https://github.com/GKalliatakis/Keras-VGG16-places365/blob/master/vgg16_places_365.py
- https://github.com/keras-team/keras-contrib