ECCV2018, sota on CARS196, 9 pages
The number of embeddings: L
The dimension of each embedding: D
D equals the number of meta-classes
Proxy NCA loss for training each embedding
The dimension of each embedding: D The number of embeddings: L
The dimension of each embedding: D The number of embeddings: L
D L
Higher than our SSR with
48 models ensemble
Discussion1:
Do we really
need attribute
to enhance feature
learning?
Samples within a meta class can
be viewed as sharing a latent
attribute. So meta classes
corresponds to randomized
attributes
Discussion2:
In hidden layers, we may expect some clusters within the dataset.
A cluster may be viewed as a meta class.
employing meta class = enforcing diversity of clustering?
Discussion3:
Encoding the original one-hot label into a sequential label.
Using L-2 loss (or KLDiv loss, etc.) for learning the embedding brings about a similar
improvement?

Deep randomized embedding

  • 1.
    ECCV2018, sota onCARS196, 9 pages
  • 3.
    The number ofembeddings: L The dimension of each embedding: D D equals the number of meta-classes Proxy NCA loss for training each embedding
  • 4.
    The dimension ofeach embedding: D The number of embeddings: L
  • 5.
    The dimension ofeach embedding: D The number of embeddings: L D L Higher than our SSR with 48 models ensemble
  • 6.
    Discussion1: Do we really needattribute to enhance feature learning? Samples within a meta class can be viewed as sharing a latent attribute. So meta classes corresponds to randomized attributes
  • 7.
    Discussion2: In hidden layers,we may expect some clusters within the dataset. A cluster may be viewed as a meta class. employing meta class = enforcing diversity of clustering? Discussion3: Encoding the original one-hot label into a sequential label. Using L-2 loss (or KLDiv loss, etc.) for learning the embedding brings about a similar improvement?