Deep Metric Learning

Learn a deep metric which can be used image retrieval , clustering.

============================

Pytorch Code for deep metric methods:

• Contrasstive Loss

• Batch-All-Loss and Batch-Hard-Loss

  2 Loss Functions in In Defense of Triplet Loss in ReID

• HistogramLoss

  Learning Deep Embeddings with Histogram Loss

• BinDevianceLoss

  Self - Modified Version with better performance

  Baseline method in BIER(Deep Metric Learning with BIER: Boosting Independent Embeddings Robustly)

• NCA Loss

  Learning a Nonlinear Embedding by Preserving Class Neighbourhood Structure -Ruslan Salakhutdinov and Geoffrey Hinton

Dataset

• Car-196

  first 98 classes as train set and last 98 classes as test set

• CUB-200-2011

  first 100 classes as train set and last 100 classes as test set

• Stanford-Online

  for the experiments, we split 59,551 images of 11,318 classes for training and 60,502 images of 11,316 classes for testing

• [In-Shop-clothes]

After downloading all the three data file, you should precess them as above, and put the directionary named DataSet in the project. We provide a script to precess CUB( Deep_Metric/DataSet/split_dataset.py ) Car and Stanford online products.

Pretrained models in Pytorch

Pre-trained VGG-16-BN

Prerequisites

• Computer with Linux or OSX
• For training, an NVIDIA GPU is strongly recommended for speed. CPU is supported but training may be slow.
• Pytorch 0.4.0

Performance of Loss:

To be clear and simple, I only provide Rank@1 on DataSets without test augment. Because, in most case, more higher the Rank@1 is, more higher the Rank@K.

In_shop_clothes result wil be updated in the near future.

Loss Function	Rank@1(%)
BinDeviance Loss	66.5
NCA Loss	61.7

Reproducing Car-196 (or CUB-200-2011) experiments

With BinDeviance Loss :

sh run_train_00.sh