Doodle It Yourself: Class Incremental Learning by Drawing a Few Sketches

Bhunia, Ayan Kumar; Gajjala, Viswanatha Reddy; Koley, Subhadeep; Kundu, Rohit; Sain, Aneeshan; Xiang, Tao; Song, Yi-Zhe

doi:10.1109/cvpr52688.2022.00233

Cited by 22 publications

(5 citation statements)

References 38 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Photorealistic image (photo) generation from free-hand sketches is still in its infancy, despite significant advances on various sketchbased vision tasks [9,11,73,84,91]. Pix2pix [41] forms the basis for most of the recent deep learning-based sketchto-photo generation frameworks (Table 1).…”

Section: Sketch-to-photo Generationmentioning

confidence: 99%

Picture that Sketch: Photorealistic Image Generation from Abstract Sketches

Koley¹,

Bhunia²,

Sain³

et al. 2023

Preprint

View full text Add to dashboard Cite

Edgemap Sketch Sketch Sketch Sketch Sketch Existing Methods Proposed Method Figure 1. (a) Set of photos generated by the proposed method. (b) While existing methods can generate faithful photos from perfectly pixel-aligned edgemaps, they fall short drastically in case of highly deformed and sparse free-hand sketches. In contrast, our autoregressive sketch-to-photo generation model produces highly photorealistic outputs from highly abstract sketches.

show abstract

Section: Sketch-to-photo Generationmentioning

confidence: 99%

Picture that Sketch: Photorealistic Image Generation from Abstract Sketches

Koley¹,

Bhunia²,

Sain³

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…-The proposed model is compared with three state-of-the-art models Subspace [30], Doodle it [34], and CEC model [35].…”

Section: Major Contributionmentioning

confidence: 99%

“…-We investigate what happens to our model if we test the impact of different loss functions, such as subspace regularizer [30] and distillation loss [34] and present the result of using each of these two loss functions. For example, The results show a decrease in the overall accuracy of around 1.4%…”

Section: Major Contributionmentioning

confidence: 99%

“…In a single session setup, it outperforms XtarNet [50] by around 3% and Attention Attractor Network [49] by about 8%. Bhunia et al [34] has presented an FSCIL method for improving the accuracy of classifying new image classes based on a few samples. The main structure has been depicted in Figure 3…”

Section: Few-shot Class Incremental Learningmentioning

confidence: 99%

“…This section compares the model with 3 states of the art models related to FSCIL and a baseline model. These models are Subspace [30], Doodle It [34], and CEC model [35]. We compare 1-shot, 5-shot and 10-shot sampling methods by considering 2 sessions for novel classes.…”

Section: Comparison Of Modelsmentioning

confidence: 99%

See 2 more Smart Citations

Intrusion Detection in IoT Network using Few-Shot Class Incremental Learning

Hosseini

View full text Add to dashboard Cite

The Internet of Things (IoT) is one of the most rapidly evolving technologies, impacting various industrial sectors. With its immense potential, IoT comes with crucial security concerns, and we face high-volume and diverse attacks that must be addressed in short periods, emphasizing the importance of utilizing intrusion detection solutions in IoT networks. In the initial stage of an intrusion detection system, when there are sufficient samples available from the known attack classes, classic network intrusion detection methods can deliver good performance. However, the learned knowledge is no longer suitable for new types of attacks with just a few samples. On the other hand, due to the limited computing ability of edge devices in distributed IoT, only a small scale of data can be used for model training. Therefore, designing a lightweight learning scheme targeting small-scale training data is essential to train or update the model more effectively in resource-constrained devices. We propose a novel model based on Few−Shot Class Incremental Learning (FSCIL) for network intrusion detection in IoT networks. This model has been used in incremental image classification tasks, and to the best of our knowledge, this is the first time that this model has been used in network intrusion detection. We compare the proposed method with some state-of-the-art methods, and experimental analysis shows that our model outperforms others.

show abstract