A Comprehensive Review of Knowledge Distillation in Computer Vision

Deep learning techniques have been demonstrated to surpass preceding cutting-edge machine learning techniques in recent years, with computer vision being one of the most prominent examples. However, deep learning models suffer from significant drawbacks when deployed in resource-constrained environments due to their large model size and high complexity. Knowledge Distillation is one of the prominent solutions to overcome this challenge. This review paper examines the current state of research on knowledge distillation, a technique for compressing complex models into smaller and simpler ones. The paper provides an overview of the major principles and techniques associated with knowledge distillation and reviews the applications of knowledge distillation in the domain of computer vision. The review focuses on the benefits of knowledge distillation, as well as the problems that must be overcome to improve its effectiveness.

Insights into the Lottery Ticket Hypothesis and Iterative Magnitude Pruning

Lottery ticket hypothesis for deep neural networks emphasizes the importance of initialization used to re-train the sparser networks obtained using the iterative magnitude pruning process. An explanation for why the specific initialization proposed by the lottery ticket hypothesis tends to work better in terms of generalization (and training) performance has been lacking. Moreover, the underlying principles in iterative magnitude pruning, like the pruning of smaller magnitude weights and the role of the iterative process, lack full understanding and explanation. In this work, we attempt to provide insights into these phenomena by empirically studying the volume/geometry and loss landscape characteristics of the solutions obtained at various stages of the iterative magnitude pruning process.

Style Description based Text-to-Speech with Conditional Prosodic Layer Normalization based Diffusion GAN

In this paper, we present a Diffusion GAN based approach (Prosodic Diff-TTS) to generate the corresponding high-fidelity speech based on the style description and content text as an input to generate speech samples within only 4 denoising steps. It leverages the novel conditional prosodic layer normalization to incorporate the style embeddings into the multi head attention based phoneme encoder and mel spectrogram decoder based generator architecture to generate the speech. The style embedding is generated by fine tuning the pretrained BERT model on auxiliary tasks such as pitch, speaking speed, emotion,gender classifications. We demonstrate the efficacy of our proposed architecture on multi-speaker LibriTTS and PromptSpeech datasets, using multiple quantitative metrics that measure generated accuracy and MOS.

Distilling Inductive Bias: Knowledge Distillation Beyond Model Compression

With the rapid development of computer vision, Vision Transformers (ViTs) offer the tantalizing prospect of unified information processing across visual and textual domains. But due to the lack of inherent inductive biases in ViTs, they require enormous amount of data for training. To make their applications practical, we introduce an innovative ensemble-based distillation approach distilling inductive bias from complementary lightweight teacher models. Prior systems relied solely on convolution-based teaching. However, this method incorporates an ensemble of light teachers with different architectural tendencies, such as convolution and involution, to instruct the student transformer jointly. Because of these unique inductive biases, instructors can accumulate a wide range of knowledge, even from readily identifiable stored datasets, which leads to enhanced student performance. Our proposed framework also involves precomputing and storing logits in advance, essentially the unnormalized predictions of the model. This optimization can accelerate the distillation process by eliminating the need for repeated forward passes during knowledge distillation, significantly reducing the computational burden and enhancing efficiency.

Single Stage Warped Cloth Learning and Semantic-Contextual Attention Feature Fusion for Virtual TryOn

Image-based virtual try-on aims to fit an in-shop garment onto a clothed person image. Garment warping, which aligns the target garment with the corresponding body parts in the person image, is a crucial step in achieving this goal. Existing methods often use multi-stage frameworks to handle clothes warping, person body synthesis and tryon generation separately or rely on noisy intermediate parser-based labels. We propose a novel single-stage framework that implicitly learns the same without explicit multi-stage learning. Our approach utilizes a novel semantic-contextual fusion attention module for garment-person feature fusion, enabling efficient and realistic cloth warping and body synthesis from target pose keypoints. By introducing a lightweight linear attention framework that attends to garment regions and fuses multiple sampled flow fields, we also address misalignment and artifacts present in previous methods. To achieve simultaneous learning of warped garment and try-on results, we introduce a Warped Cloth Learning Module. WCLM uses segmented warped garments as ground truth, operating within a single-stage paradigm. Our proposed approach significantly improves the quality and efficiency of virtual try-on methods, providing users with a more reliable and realistic virtual try-on experience. We evaluate our method on the VITON dataset and demonstrate its state-of-the-art performance in terms of both qualitative and quantitative metrics.

Transformer-based Nonlinear Equalization for DP-16QAM Coherent Optical Communication Systems

Compensating for nonlinear effects using digital signal processing (DSP) is complex and computationally expensive in long-haul optical communication systems due to intractable interactions between Kerr nonlinearity, chromatic dispersion (CD), and amplified spontaneous emission (ASE) noise from inline amplifiers. The application of machine learning architectures has demonstrated promising advancements in enhancing transmission performance through the mitigation of fiber nonlinear effects. In this paper, we apply a Transformer-based model to dual-polarisation (DP)-16QAM coherent optical communication systems. We test the performance of the proposed model for different values of fiber lengths and launched optical powers and show improved performance compared to the state-of-the-art digital backpropagation (DBP) algorithm, fully connected neural network (FCNN) and bidirectional long short term memory (BiLSTM) architecture.

IIITD-20K: Dense captioning for Text-Image ReID

Text-to-Image (T2I) ReID has attracted a lot of attention in the recent past. CUHK-PEDES, RSTPReid and ICFG-PEDES are the three available benchmarks to evaluate T2I ReID methods. RSTPReid and ICFG-PEDES comprise of identities from MSMT17 but due to limited number of unique persons, the diversity is limited. On the other hand, CUHK-PEDES comprises of 13,003 identities but has relatively shorter text description on average. Further, these datasets are captured in a restricted environment with limited number of cameras. In order to further diversify the identities and provide dense captions, we propose a novel dataset called IIITD-20K. IIITD-20K comprises of 20,000 unique identities captured in the wild and provides a rich dataset for text-to-image ReID. With a minimum of 26 words for a description, each image is densely captioned. We further synthetically generate images and fine-grained captions using Stable-diffusion and BLIP models trained on our dataset. We perform elaborate experiments using state-of-art text-to-image ReID models and vision-language pre-trained models and present a comprehensive analysis of the dataset. Our experiments also reveal that synthetically generated data leads to a substantial performance improvement in both same dataset as well as cross dataset settings. Our dataset is available at https://bit.ly/3pkA3Rj.

A Language-Guided Benchmark for Weakly Supervised Open Vocabulary Semantic Segmentation

Increasing attention is being diverted to data-efficient problem settings like Open Vocabulary Semantic Segmentation (OVSS) which deals with segmenting an arbitrary object that may or may not be seen during training. The closest standard problems related to OVSS are Zero-Shot and Few-Shot Segmentation (ZSS, FSS) and their Cross-dataset variants where zero to few annotations are needed to segment novel classes. The existing FSS and ZSS methods utilize fully supervised pixel-labelled seen classes to segment unseen classes. Pixel-level labels are hard to obtain, and using weak supervision in the form of inexpensive image-level labels is often more practical. To this end, we propose a novel unified weakly supervised OVSS pipeline that can perform ZSS, FSS and Cross-dataset segmentation on novel classes without using pixel-level labels for either the base (seen) or the novel (unseen) classes in an inductive setting. We propose Weakly-Supervised Language-Guided Segmentation Network (WLSegNet), a novel language-guided segmentation pipeline that i) learns generalizable context vectors with batch aggregates (mean) to map class prompts to image features using frozen CLIP (a vision-language model) and ii) decouples weak ZSS/FSS into weak semantic segmentation and Zero-Shot segmentation. The learned context vectors avoid overfitting on seen classes during training and transfer better to novel classes during testing. WLSegNet avoids fine-tuning and the use of external datasets during training. The proposed pipeline beats existing methods for weak generalized Zero-Shot and weak Few-Shot semantic segmentation by 39 and 3 mIOU points respectively on PASCAL VOC and weak Few-Shot semantic segmentation by 5 mIOU points on MS COCO. On a harder setting of 2-way 1-shot weak FSS, WLSegNet beats the baselines by 13 and 22 mIOU points on PASCAL VOC and MS COCO, respectively.

Knowledge Distillation in Vision Transformers: A Critical Review

In Natural Language Processing (NLP), Transformers have already revolutionized the field by utilizing an attention-based encoder-decoder model. Recently, some pioneering works have employed Transformer-like architectures in Computer Vision (CV) and they have reported outstanding performance of these architectures in tasks such as image classification, object detection, and semantic segmentation. Vision Transformers (ViTs) have demonstrated impressive performance improvements over Convolutional Neural Networks (CNNs) due to their competitive modelling capabilities. However, these architectures demand massive computational resources which makes these models difficult to be deployed in the resource-constrained applications. Many solutions have been developed to combat this issue, such as compressive transformers and compression functions such as dilated convolution, min-max pooling, 1D convolution, etc. Model compression has recently attracted considerable research attention as a potential remedy. A number of model compression methods have been proposed in the literature such as weight quantization, weight multiplexing, pruning and Knowledge Distillation (KD). However, techniques like weight quantization, pruning and weight multiplexing typically involve complex pipelines for performing the compression. KD has been found to be a simple and much effective model compression technique that allows a relatively simple model to perform tasks almost as accurately as a complex model. This paper discusses various approaches based upon KD for effective compression of ViT models. The paper elucidates the role played by KD in reducing the computational and memory requirements of these models. The paper also presents the various challenges faced by ViTs that are yet to be resolved.

KL Regularized Normalization Framework for Low Resource Tasks

Large pre-trained models, such as Bert, GPT, and Wav2Vec, have demonstrated great potential for learning representations that are transferable to a wide variety of downstream tasks . It is difficult to obtain a large quantity of supervised data due to the limited availability of resources and time. In light of this, a significant amount of research has been conducted in the area of adopting large pre-trained datasets for diverse downstream tasks via fine tuning, linear probing, or prompt tuning in low resource settings. Normalization techniques are essential for accelerating training and improving the generalization of deep neural networks and have been successfully used in a wide variety of applications. A lot of normalization techniques have been proposed but the success of normalization in low resource downstream NLP and speech tasks is limited. One of the reasons is the inability to capture expressiveness by rescaling parameters of normalization. We propose KullbackLeibler(KL) Regularized normalization (KL-Norm) which make the normalized data well behaved and helps in better generalization as it reduces over-fitting, generalises well on out of domain distributions and removes irrelevant biases and features with negligible increase in model parameters and memory overheads. Detailed experimental evaluation on multiple low resource NLP and speech tasks, demonstrates the superior performance of KL-Norm as compared to other popular normalization and regularization techniques.