Training Model by Knowledge Distillation for Image-text Matching Use knowledge distillation method to compress pre-trained models in Image-Text matching tasks.Design lightweight models and use knowledge distillation methods to achieve better results for previously ineffective models after training.
Hai LiuXingxing YaoXiangyu KongPublished in: ICAICE (2023)
Keyphrases
- computer vision
- image matching
- prior knowledge
- probabilistic model
- image features
- statistical model
- image segmentation
- image set
- lightweight
- object detectors
- test images
- parametric models
- pre trained
- input image
- energy function
- feature points
- matching algorithm
- pairwise
- feature matching
- active appearance models
- single image
- neural network
- supervised learning
- learning algorithm