Awesome

DeltaGAN-Few-Shot-Image-Generation

Code for our paper "DeltaGAN: Towards Diverse Few-shot Image Generation with Sample-Specific Delta". [pdf]

Created by Yan Hong, Li Niu*, Jianfu Zhang, Liqing Zhang.

Accepted by ECCV2022.

Citation

If you find our work useful in your research, please consider citing:

@inproceedings{HongDeltaGAN,
  title={DeltaGAN: Towards Diverse Few-shot Image Generation with Sample-Specific Delta},
  author={Hong, Yan and Niu, Li and Zhang, Jianfu and Zhang, Liqing},
  booktitle={ECCV},
  year={2022}
}

Introduction

Few-shot image generation aims at generating images for a new category with only a few images, which can make fast adaptation to a new category especially for those newly emerging categories or long-tail categories. Few-shot image generation can be used for data augmentation, which benefits a wide range of downstream category-aware tasks like few-shot classification.Several state-of-the-art works have yielded impressive results, but the diversity is still limited. In this work, we propose a novel Delta Generative Adversarial Network (DeltaGAN), which consists of a reconstruction subnetwork and a generation subnetwork. The reconstruction subnetwork captures intra-category transformation, \emph{i.e.}, delta, between same-category pairs. The generation subnetwork generates sample-specific delta for an input image, which is combined with this input image to generate a new image within the same category. Besides, an adversarial delta matching loss is designed to link the above two subnetworks together. Extensive experiments on six benchmark datasets demonstrate the effectiveness of our proposed method.

Comparison Visualization

More Visualization

Experiments

Hardware& Software Dependency

• Hardware

  a single GPU or multiple GPUs

• Software

  Tensorflow-gpu (version >= 1.7)

  Opencv

  scipy

• Click here to view detailed software dependency

Datasets Preparation

• The Download links can be found here

• Emnist

  Categories/Samples: 38/ 106400

  Split: 28 seen classes, 10 unseen classes

• VGGFace

  Categories/Samples: 2299/ 229900

  Split: 1802 seen classes, 497 unseen classes

• Flowers

  Categories/Samples:** 102/ 8189

  Split: 85 seen classes, 17 unseen classes

• Animal Faces

  Categories/Samples: 149/ 214105

  Split: 119 seen classes, 30 unseen classes

• NABirds

  Categories/Samples: 555/ 48527

  Split: 444 seen classes, 111 unseen classes

• Foods

  Categories/Samples: 256/ 31395

  Split: 224 seen classes, 32 unseen classes

Baselines

Few-shot Image Generation

• FIGR: Few-shot Image Generation with Reptile paper code

• Few-shot Generative Modelling with Generative Matching Networks paper code

• DAWSON: A do- main adaptive few shot generation framework paper code

• Data Augmentation Generative Adversarial Networks paper code

• F2GAN: Fusing-and-Filling GAN for Few-shot Image Generation paper code

• Matchinggan: Matching-Based Few-Shot Image Generationpaper code

• LoFGAN: Fusing Local Representations for Few-shot Image Generationpapercode

Few-shot Image Classification

• Matching Networks for One Shot Learning paper code

• Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks paper code

• Learning to Compare: Relation Network for Few-Shot Learning paper code

• DPGN: Distribution Propagation Graph Network for Few-shot Learning paper code

• Meta-Transfer Learning for Few-Shot Learning paper code

• Cross-Domain Few-Shot Classification via Learned Feature-Wise Transformation paper code

• Delta-encoder: an effective sample synthesis methodfor few-shot object recognitio papercode

Getting Started

Installation

1.Clone this repository.

git clone https://github.com/bcmi/DeltaGAN-Few-Shot-Image-Generation.git

2.Create python environment for DeltaGAN via pip.

pip install -r requirements.txt

Data Preprecessing

1. Dwonloading the datasets, obtaining the path of corresponding datasets 'dataroot', and setting the path for preprocessed dataset 'storepath'

2. Runing the script, obtaining list [num_categories, num_samples_each_category, image_width, image_height, image_channel]

python data_preparation.py --dataroot --storepath --image_width 96 --channel 3 

• for Emnist: --image_width 28, --channel 1
• for other datasets: --image_width 96, --channel 3

3. Setting the storepath in the data_with_matchingclassifier.py for each dataset. For example, replacing the path in class EmnistDAGANDataset with the storepath of your preprocessed Emnist data.

4. If your need run the other datasets except for our selected five datasets, you can also follow the above data preprocessing

Training

1.Train on EMNIST dataset

python train_dagan_with_matchingclassifier.py --dataset emnist --image_width 28 --batch_size 20  --experiment_title MMF2GAN/emnist1way3shot   --selected_classes 1 --support_number 2  --loss_G 1 --loss_D 1 --loss_CLA 1  --loss_recons_B 1 --loss_matching_G 0.01 --loss_matching_D 1 --loss_sim 1 

2.Train on VGGFce dataset

python train_dagan_with_matchingclassifier.py --dataset vggface --image_width 96 --batch_size 20  --experiment_title MMF2GAN/vggface1way3shot   --selected_classes 1 --support_number 3  --loss_G 1 --loss_D 1 --loss_CLA 1  --loss_recons_B 1 --loss_matching_G 0.01 --loss_matching_D 1 --loss_sim 1 

3.Train on Flowers dataset

python train_dagan_with_matchingclassifier.py --dataset flowers --image_width 96 --batch_size 20  --experiment_title MMF2GAN/flowers1way3shot   --selected_classes 1 --support_number 2  --loss_G 1 --loss_D 1 --loss_CLA 1  --loss_recons_B 1 --loss_matching_G 0.01 --loss_matching_D 1 --loss_sim 1 

4.Train on Animal Faces dataset

python train_dagan_with_matchingclassifier.py --dataset animals --image_width 96 --batch_size 20 --generation_layers 4 --generator_inner_layers 2 --num_generations 32 --experiment_title animals1way2shot42layersLikeStargenV2allNoSharedOneDUniformSameDimensionLowerx1MSDCombinedDx1x2  --selected_classes 1 --support_number 2  --loss_G 0.5 --loss_D 1 --loss_KL 0 --loss_CLA 1 --loss_FSL 10 --loss_recons_B 1 --loss_matching_G 10 --loss_matching_D 0.1 --loss_sim 0 --z_dim 128  

Trained Model

• EMNIST: emnist_trained model, extracted code: 1234.

• VGGFace: vggface_trained model, extracted code: 1234.

• Flowers: flowers_trained model, extracted code: 1234.

• Animal Faces: animals_trained model, extracted code: 1234.

• NAbirds: nabirds_trained model, extracted code: 1234.

Evaluation from three aspects including GAN metrics, low-data classification, and few-shot classification.

1. Visualizing the generated images based on trained models, the generated images are stored in the path '--experiment_title'

EMNIST generated images

python test_dagan_with_matchingclassifier_for_generation.py  --is_training 0 --is_all_test_categories 1 --is_generation_for_classifier 1  --general_classification_samples 10 --dataset emnist --image_width 28  --batch_size 30  --num_generations 128 --experiment_title EVALUATION_Augmented_emnist_DeltaGAN --selected_classes 1 --support_number 3   --restore_path   ./trained_models/emnist/  --continue_from_epoch 100

VGGFace generated images

python test_dagan_with_matchingclassifier_for_generation.py  --is_training 0 --is_all_test_categories 1 --is_generation_for_classifier 1  --general_classification_samples 10 --dataset vggface --image_width 96  --batch_size 30  --num_generations 128 --experiment_title EVALUATION_Augmented_vggface_DeltaGAN --selected_classes 1 --support_number 3   --restore_path  path  ./trained_models/vggface/  --continue_from_epoch 100

Flowers generated images

python test_dagan_with_matchingclassifier_for_generation.py  --is_training 0 --is_all_test_categories 1 --is_generation_for_classifier 1  --general_classification_samples 10 --dataset flowers --image_width 96  --batch_size 30  --num_generations 128 --experiment_title EVALUATION_Augmented_flowers_DeltaGAN --selected_classes 1 --support_number 3   --restore_path path   ./trained_models/flowers/  --continue_from_epoch  100

Animal Faces generated images

python test_dagan_with_matchingclassifier_for_generation.py  --is_training 0 --is_all_test_categories 1 --is_generation_for_classifier 1  --general_classification_samples 10 --dataset animals --generation_layers 4 --generator_inner_layers 2 --image_width 96 --batch_size 30  --num_generations 100 --experiment_title EVALUATION_Augmented_animals --selected_classes 1 --support_number 3 --loss_G 1 --loss_D 1 --loss_KL 0 --loss_CLA 1 --loss_recons_B 1 --loss_FSL 10 --loss_sim 1 --loss_matching_D 0.1 --loss_matching_G 10 --z_dim 128  --restore_path   --continue_from_epoch 

NABirds generated images

python test_dagan_with_matchingclassifier_for_generation.py  --is_training 0 --is_all_test_categories 1 --is_generation_for_classifier 1  --general_classification_samples 10 --dataset nabirds --generation_layers 4 --generator_inner_layers 2 --image_width 96 --batch_size 30  --num_generations 100 --experiment_title EVALUATION_Augmented_animals --selected_classes 1 --support_number 3 --loss_G 1 --loss_D 1 --loss_KL 0 --loss_CLA 1 --loss_recons_B 1 --loss_FSL 10 --loss_sim 1 --loss_matching_D 0.1 --loss_matching_G 10 --z_dim 128  --restore_path   --continue_from_epoch 

2. Testing the GAN metrics including IS, FID, and IPIPS for generated images, which is suitable for RGB 3-channel images like VGGFace, Flowers, Animal Faces, and NABirds datasets.

VGGFace GAN metrics

python GAN_metrcis_FID_IS_LPIPS.py  --dataroot_real ./EVALUATION/Augmented/vggface/DeltaGAN/visual_outputs_realimages/ --dataroot_fake  ./EVALUATION/Augmented/vggface/DeltaGAN/visual_outputs_forquality/  --image_width 128 --image_channel 3 --augmented_support 100  --dir ./EVALUATION/Augmented/vggface/DeltaGAN/visual_outputs_forquality/ --out ./EVALUATION/Augmented/vggface/DeltaGAN/GAN_METRICS.txt 

Flowers GAN metrics

python GAN_metrcis_FID_IS_LPIPS.py  --dataroot_real ./EVALUATION/Augmented/flowers/DeltaGAN/visual_outputs_realimages/ --dataroot_fake  ./EVALUATION/Augmented/flowers/DeltaGAN/visual_outputs_forquality/  --image_width 128 --image_channel 3 --augmented_support 100  --dir ./EVALUATION/Augmented/flowers/DeltaGAN/visual_outputs_forquality/ --out ./EVALUATION/Augmented/flowers/DeltaGAN/GAN_METRICS.txt 

Animal Faces GAN metrics

python GAN_metrcis_FID_IS_LPIPS.py  --dataroot_real ./EVALUATION/Augmented/animals/DeltaGAN/visual_outputs_realimages/ --dataroot_fake  ./EVALUATION/Augmented/animals/DeltaGAN/visual_outputs_forquality/  --image_width 128 --image_channel 3 --augmented_support 100  --dir ./EVALUATION/Augmented/animals/DeltaGAN/visual_outputs_forquality/ --out ./EVALUATION/Augmented/animals/DeltaGAN/GAN_METRICS.txt 

NABirds GAN metrics

python GAN_metrcis_FID_IS_LPIPS.py  --dataroot_real ./EVALUATION/Augmented/nabirds/DeltaGAN/visual_outputs_realimages/ --dataroot_fake  ./EVALUATION/Augmented/nabirds/DeltaGAN/visual_outputs_forquality/  --image_width 128 --image_channel 3 --augmented_support 100  --dir ./EVALUATION/Augmented/nabirds/DeltaGAN/visual_outputs_forquality/ --out ./EVALUATION/Augmented/nabirds/DeltaGAN/GAN_METRICS.txt 

3. Testing the classification in low-data setting with augmented images.

take EMNIST as example, low-data classification with augmented images generated from our trained model

3.1. Gnerating augmented images using three conditional images

python test_dagan_with_matchingclassifier_for_generation.py  --is_training 0 --is_all_test_categories 1 --is_generation_for_classifier 1  --general_classification_samples 10 --dataset emnist --image_width 28  --batch_size 30  --num_generations 512 --experiment_title EVALUATION_Augmented_emnist_DeltaGAN --selected_classes 1 --support_number 3   --restore_path path ./trained_models/emnist/ --continue_from_epoch 100

3.2. Preparing generated images: the generated images are stored in the 'storepath/visual_outputs_forclassifier' and setting the storepath for preprocessed data, running below script

python data_preparation.py --dataroot storepath/visual_outputs_forclassifier  --storepath --image_width 28 --channel 1 

3.3. Replacing the datapath in data_with_matchingclassifier_for_quality_and_classifier.py with the storepath for preprocessed data.

3.4. Running the script for low-data classification.

train_classifier_with_augmented_images.py --dataset emnist  --selected_classes testing_categories  --batch_size 16 --classification_total_epoch 50  --experiment_title AugmentedLowdataClassifier_emnist  --image_width 28  --image_height 28 --image_channel 1

--selected_classes: the number of total testing categories

4. Testing the classification in few-shot setting with augmented images.

take EMNIST as example, NwayKshot classification with augmented images generated from our trained model

4.1. Gnerating augmented images using Kshot conditional images

python test_dagan_with_matchingclassifier_for_generation.py  --is_training 0 --is_all_test_categories 1 --is_generation_for_classifier 1  --general_classification_samples 10 --dataset emnist --image_width 28  --batch_size 30  --num_generations 128 --experiment_title EVALUATION_Augmented_emnist_emnist --selected_classes 1 --support_number K   --restore_path path ./trained_models/emnist/ --continue_from_epoch 100

setting the '--support_number' as K.

4.2. Preprocessing the generated images

python data_preparation.py --dataroot ./EVALUATION/Augmented/emnist/DeltaGAN/visual_outputs_forclassifier  --storepath ./EVALUATION/Augmented/emnist/DeltaGAN/  --image_width 28 --channel 1 

4.3. Replacing the datapath in data_with_matchingclassifier_for_quality_and_classifier.py with ./EVALUATION/Augmented/emnist/DeltaGAN/emnist.npy.

4.4. Running the script for few-shot classification.

train_classifier_with_augmented_images.py --dataset emnist  --selected_classes N  --batch_size 16 --classification_total_epoch 50  --experiment_title AugmentedFewshotClassifier_emnist  --image_width 28  --image_height 28 --image_channel 1

setting the '--selected_classes' as N.

Results

To view more clear results, please click the belowing tables.

GAN metrics of Generated Images

Few-shot Image Classification

Acknowledgement

Some of the codes are built upon DAGAN. Thanks them for their great work!

If you get any problems or if you find any bugs, don't hesitate to comment on GitHub or make a pull request!

DeltaGAN is freely available for non-commercial use, and may be redistributed under these conditions. For commercial queries, please drop an e-mail. We will send the detail agreement to you.