hedged_instance_embedding
embedding method for measuring uncertainty
Introduction(from abstract)
Often the distance between points is used as a proxy for match confidence. However, this can fail to represent uncertainty arising when the input is ambiguous, e.g., due to occlusion or blurriness
This work addresses this issue and explicitly models the uncertainty by hedging the location of each input in the embedding space.
Dependency
- numpy 1.16.3
- tensorflow 2.0
- tfp-nightly 0.7.0
- matplotlib 3.0.3
- sklearn 0.21.1
Usage
- train
python train.py [option]- visualize
python visualize.py [option]Method
soft constrative loss: point embedding loss
In paper, see 2 - 1 - (3)
VIB loss: hedged instance embedding loss
In paper, see 2 - 3 - (8)
Result: Point Emb VS HIB Emb
- Input image 80: clean vs occlusion
- Input image 58: clean vs occlusion
