Scribd
Upload
19 views

Paper Review of Five Machine Vision Topics

Uploaded by

sashidhar avuthu
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
19 views

Paper Review of Five Machine Vision Topics

Uploaded by

sashidhar avuthu
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

Paper Review of Five Machine Vision Topics

Sashidhar Reddy Avuthu


Sa2220

1. Introduction of Background
The field of machine vision has seen exponential growth with the integration of deep learning,
providing breakthroughs across diverse applications like object detection, image restoration,
segmentation, facial recognition, and scene understanding. This review covers five papers
highlighting different aspects of machine vision, exploring methodologies, proposed solutions,
and their contributions to real-world scenarios.

2. Related Works for Quick Comparison


- Object Detection Using Deep Learning, CNNs, and Vision Transformers has built upon
foundational works like YOLO and Faster R-CNN, extending capabilities through Vision
Transformers for more accurate object representation.
- Image Restoration via Neural Networks and GANs relates to traditional image processing
methods such as wavelet transforms but moves beyond through GANs' capacity to generate
visually plausible content.
- Semantic Segmentation of Urban Scenes Using Deep Networks contrasts with earlier
segmentation techniques by employing a modified U-Net with dense connections, addressing
challenges in urban scene complexity.
- Facial Recognition Using Sparse Representation and Deep Learning enhances previous
sparse representation models with deep learning layers for robust occlusion handling.
- Scene Understanding with Self-Supervised Learning leverages self-supervised
frameworks, contrasting with fully supervised methods that demand extensive labeled datasets.

3. Proposed Methods and Results


- Object Detection: This paper proposed combining CNNs and Vision Transformers, enabling
models to capture long-range dependencies and achieve state-of-the-art detection accuracy on
public benchmarks like COCO.
- Results: Significant improvement in object localization and classification metrics compared to
traditional CNN-based models.
- Image Restoration with GANs: The proposed approach utilized a GAN-based architecture for
denoising and inpainting, generating clearer images than conventional methods.
- Results: Achieved notable reductions in noise levels with visual quality improvements on
benchmark datasets like ImageNet.
- Semantic Segmentation: The improved U-Net architecture, coupled with dense connections,
optimized pixel-level classification tasks in urban scenes.
- Results: Outperformed baseline models in terms of accuracy and robustness on the
Cityscapes dataset.
- Facial Recognition: The hybrid model integrated sparse representation with deep
convolutional layers for enhanced recognition capabilities.
- Results: High recognition accuracy in datasets with occluded faces, outperforming models
that used only sparse representation or deep learning.
- Scene Understanding with Self-Supervised Learning: By leveraging contrastive learning,
this method reduced reliance on labeled data for scene feature extraction.
- Results: Effective cross-domain generalization on several scene understanding benchmarks,
albeit with a slight performance gap compared to supervised models.

4. Analysis and Summarization of Pros and Cons


Object Detection Using Deep Learning, CNNs, and Vision Transformers:
- Pros: Superior object representation; scalable model structure.
- Cons: Computationally intensive; real-time deployment challenges.
Image Restoration via Neural Networks and GANs:
- Pros: High visual fidelity for image restoration; adaptable to various image degradation types.
- Cons: Training complexity and stability issues.
Semantic Segmentation of Urban Scenes:
- Pros: Robust in complex urban settings; accurate segmentation.
- Cons: High memory and computational costs; slower inference speed.
Facial Recognition Using Sparse Representation and Deep Learning:
- Pros: Effective against occlusions; high recognition accuracy.
- Cons: Limited scalability to real-time systems.
Scene Understanding with Self-Supervised Learning:
- Pros: Cost-effective data labeling; good cross-domain performance.
- Cons: Lower performance on fine-grained details compared to supervised models.

References
1. Amjoud, Ayoub Benali, and Mustapha Amrouch. "Object detection using deep learning, CNNs and
vision transformers: A review." IEEE Access 11 (2023): 35479-35516.
2. Rama, P., et al. "Advancement in Image Restoration Through GAN-based Approach." 2024 15th
International Conference on Computing Communication and Networking Technologies (ICCCNT). IEEE,
2024.
3. Li, Yanyi, Jian Shi, and Yuping Li. "Real-Time Semantic Understanding and Segmentation of Urban
Scenes for Vehicle Visual Sensors by Optimized DCNN Algorithm." Applied Sciences 12.15 (2022): 7811.
4. Wright, John, et al. "Robust face recognition via sparse representation." IEEE transactions on pattern
analysis and machine intelligence 31.2 (2008): 210-227.
5. Jiang, Huaizu, et al. "Self-supervised relative depth learning for urban scene understanding."
Proceedings of the european conference on computer vision (eccv). 2018.

You might also like