[ICML 2025] Provably Improving Generalization of Few-shot models with synthetic data

This is the implementation of full version of paper Provably Improving Generalization of Few-shot models with synthetic data.

This code is heavily adapted from DataDream codebase: https://github.com/ExplainableML/DataDream. Thanks DataDream authors for their great work!

The codebase for lightweight version can be found at: https://github.com/cuonglannguyen/ProtoAug_lightweight

Preliminary Setup

Before we start, few-shot real data should be formed in the following way. Assuming we use 16-shot, each data files should be located in the path data/$DATASET/real_train_fewshot/seed$SEED_NUMBER/$CLASS_NAME/$FILE. The list of CLASS_NAME For each DATASET can be found in DataDream/util_data.py file.

For instance, the ImageNet 16-shot data should be stored as follows:

📂 data
|_📂 imagenet
  |_📂 real_train_fewshot
    |_📂 seed0
      |_📂 abacus
        |_📄 n02666196_17944.JPEG
        |_📄 n02666196_10754.JPEG
        |_📄 n02666196_10341.JPEG
        |_📄 n02666196_26262.JPEG
        |_📄 n02666196_16203.JPEG
        |_📄 n02666196_15765.JPEG
        |_📄 n02666196_16339.JPEG
        |_📄 n02666196_7225.JPEG
        |_📄 n02666196_13227.JPEG
        |_📄 n02666196_19345.JPEG
        |_📄 n02666196_19170.JPEG
        |_📄 n02666196_9008.JPEG
        |_📄 n02666196_20311.JPEG
        |_📄 n02666196_17676.JPEG
        |_📄 n02666196_16649.JPEG
      |_📂 clothes iron
      |_📂 great white shark
      |_📂 goldfish
      |_📂 tench
      ...

Step

1. Install the necessary dependencies in requirements.txt.
2. DataDream: Follow the instructions in the DataDream folder.
3. Dataset Generation: Follow the instructions in the generate folder.
4. Train Classifier: Follow the instructions in the classify folder.