GAN Slimming: All-in-One GAN Compression by a Unified Optimization Framework

Wang, Haotao; Gui, Shupeng; Yang, Haichuan; Liu, Ji; Wang, Zhangyang

doi:10.1007/978-3-030-58548-8_4

Cited by 74 publications

(62 citation statements)

References 38 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The mainstream of conditional GAN compression [26], [27], [28], [29], [30], [31], [32], [33], [49] exploits KD, transferring the knowledge of teacher generator to student generator. Besides KD, several works [14], [16], [19] exploit network pruning. In addition, there exist several works [24], [25] developing evolutionary compression with inferior results compared with KD-based approaches.…”

Section: Related Workmentioning

confidence: 99%

“…To make GANs easy to be deployed on computational resource limited devices, extensive works have been proposed to obtain lightweight GANs. The mainstreaming approach is to inherit the model compression techniques developed for image-classification task to compress GANs, such as weight pruning [14], weight quantization [15], channel pruning [16], [17], [18], [19], lightweight GAN architecture search/design [20], [21], [22], [23], evolutionary compression [24], [25], and knowledge distillation (KD) [26], [27], [28], [29], [30], [31], [32], [33]. However, most of the above works focus on compressing conditional (cycle) GANs for image-to-image generation tasks, scarce works have been proposed for compressing vanilla GANs except recent works [8], [17], [18], [27], [34].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

DGL-GAN: Discriminator Guided Learning for GAN Compression

Tian¹,

Shen²,

Tao³

et al. 2021

Preprint

View full text Add to dashboard Cite

Generative Adversarial Networks (GANs) with high computation costs, e.g., BigGAN and StyleGAN2, have achieved remarkable results in synthesizing high resolution and diverse images with high fidelity from random noises. Reducing the computation cost of GANs while keeping generating photo-realistic images is an urgent and challenging field for their broad applications on computational resource-limited devices. In this work, we propose a novel yet simple Discriminator Guided Learning approach for compressing vanilla GAN, dubbed DGL-GAN. Motivated by the phenomenon that the teacher discriminator may contain some meaningful information, we transfer the knowledge merely from the teacher discriminator via the adversarial function. We show DGL-GAN is valid since empirically, learning from the teacher discriminator could facilitate the performance of student GANs, verified by extensive experimental findings. Furthermore, we propose a two-stage training strategy for training DGL-GAN, which can largely stabilize its training process and achieve superior performance when we apply DGL-GAN to compress the two most representative large-scale vanilla GANs, i.e., StyleGAN2 and BigGAN. Experiments show that DGL-GAN achieves state-of-the-art (SOTA) results on both StyleGAN2 (FID 2.92 on FFHQ with nearly 1/3 parameters of StyleGAN2) and BigGAN (IS 93.29 and FID 9.92 on ImageNet with nearly 1/4 parameters of BigGAN) and also outperforms several existing vanilla GAN compression techniques. Moreover, DGL-GAN is also effective in boosting the performance of original uncompressed GANs, original uncompressed StyleGAN2 boosted with DGL-GAN achieves FID 2.65 on FFHQ, which achieves a new state-of-the-art performance. Code and models are available at https://github.com/yuesongtian/DGL-GAN.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

DGL-GAN: Discriminator Guided Learning for GAN Compression

Tian¹,

Shen²,

Tao³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…Although it is theoretically possible to employ this method for low bitwidth inference in generative models, the non-uniform scheme utilized complicates deployment of this framework on edge devices. Perhaps, the most relevant method was described in [26]. The authors applied uniform quantization to both weights and activations, combining quantization with pruning and knowledge distillation in a unified optimization framework.…”

Section: Quantization Of Generative Adversarial Networkmentioning

confidence: 99%

“…Hence, we make use of quantization-aware training to further reduce the quality degradation associated with quantization. Similarly to [26], we employ a sum of adversarial and reconstruction losses for quantization-aware training:…”

Section: Quantization-aware Trainingmentioning

confidence: 99%

Quantization of Generative Adversarial Networks for Efficient Inference: a Methodological Study

Andreev¹,

Fritzler²,

Vetrov³

2021

Preprint

View full text Add to dashboard Cite

Generative adversarial networks (GANs) have an enormous potential impact on digital content creation, e.g., photo-realistic digital avatars, semantic content editing, and quality enhancement of speech and images. However, the performance of modern GANs comes together with massive amounts of computations performed during the inference and high energy consumption. That complicates, or even makes impossible, their deployment on edge devices. The problem can be reduced with quantization-a neural network compression technique that facilitates hardware-friendly inference by replacing floating-point computations with low-bit integer ones. While quantization is well established for discriminative models, the performance of modern quantization techniques in application to GANs remains unclear. GANs generate content of a more complex structure than discriminative models, and thus quantization of GANs is significantly more challenging. To tackle this problem, we perform an extensive experimental study of state-of-art quantization techniques on three diverse GAN architectures, namely StyleGAN, Self-Attention GAN, and CycleGAN. As a result, we discovered practical recipes that allowed us to successfully quantize these models for inference with 4/8-bit weights and 8-bit activations while preserving the quality of the original full-precision models.

show abstract

“…However, they fell into the trap of manual pre-defined search space and huge search costs. The prune-based methods [38,31,57,61] directly pruned a lightweight generator architecture from the original generator architecture. However, these works failed to take discriminator pruning into account, which would seriously destroy the Nash equilibrium between the generator and the discriminator.…”

Section: Related Workmentioning

confidence: 99%

Revisiting Discriminator in GAN Compression: A Generator-discriminator Cooperative Compression Scheme

Li¹,

Wu²,

Xiao³

et al. 2021

Preprint

View full text Add to dashboard Cite

Recently, a series of algorithms have been explored for GAN compression, which aims to reduce tremendous computational overhead and memory usages when deploying GANs on resource-constrained edge devices. However, most of the existing GAN compression work only focuses on how to compress the generator, while fails to take the discriminator into account. In this work, we revisit the role of discriminator in GAN compression and design a novel generator-discriminator cooperative compression scheme for GAN compression, termed GCC. Within GCC, a selective activation discriminator automatically selects and activates convolutional channels according to a local capacity constraint and a global coordination constraint, which help maintain the Nash equilibrium with the lightweight generator during the adversarial training and avoid mode collapse. The original generator and discriminator are also optimized from scratch, to play as a teacher model to progressively refine the pruned generator and the selective activation discriminator. A novel online collaborative distillation scheme is designed to take full advantage of the intermediate feature of the teacher generator and discriminator to further boost the performance of the lightweight generator. Extensive experiments on various GAN-based generation tasks demonstrate the effectiveness and generalization of GCC. Among them, GCC contributes to reducing 80% computational costs while maintains comparable performance in image translation tasks. Our code and models are available at: https://github.com/SJLeo/GCC * This work was done when Shaojie Li was a intern at Bytedance Inc.

show abstract

GAN Slimming: All-in-One GAN Compression by a Unified Optimization Framework

Cited by 74 publications

References 38 publications

DGL-GAN: Discriminator Guided Learning for GAN Compression

DGL-GAN: Discriminator Guided Learning for GAN Compression

Quantization of Generative Adversarial Networks for Efficient Inference: a Methodological Study

Revisiting Discriminator in GAN Compression: A Generator-discriminator Cooperative Compression Scheme

Contact Info

Product

Resources

About