In recent years, vision transformers (ViTs) have made significant breakthroughs in computer vision and have demonstrated great potential in large-scale models. However, the quantization methods for convolutional neural network models do not perform well on ViTs models, leading to a significant decrease in accuracy when applied to ViTs models. We extend the quantization parameter optimization method based on the Hessian matrix and apply it to the quantization of the LayerNorm module in ViT models. This approach reduces the impact of quantization on task accuracy for the LayerNorm module and enables more comprehensive quantization of ViT models. To achieve fast quantization of ViTs models, we propose a quantization framework specifically designed for ViTs models: Hessian matrix–aware post-training quantization for vision transformers (HAPTQ). The experimental results on various models and datasets demonstrate that our HAPTQ method, after quantizing the LayerNorm module of various ViT models, can achieve lossless quantization (with an accuracy drop of less than 1%) in ImageNet classification tasks. Specifically, the HAPTQ method achieves 85.81% top-1 accuracy on the ViT-L model.
Combustible gases exhibit high absorption capabilities in the mid-infrared spectrum, making mid-infrared imagery a promising trend for visualizing future combustible gas leakage. However, due to the inconspicuous features of combustible gases in mid-infrared images and their susceptibility to interference from other moving objects, the detection error rate remains high. To address this, a combustible gas leakage detection algorithm that integrates foreground region detection with semantic segmentation has been devised. Initially, average background is computed using confidence intervals and recursive methods, coupled with cumulative mean and variance subimages for foreground denoising. Subsequently, adaptive threshold segmentation based on grayscale histograms is employed to identify all moving foreground regions. Furthermore, a multimodal fusion semantic segmentation network, GasLeakNet, is introduced to eliminate non-gas foreground targets. Ultimately, the proposed method achieves gas region extraction and visualization through the fusion of foreground region extraction and semantic segmentation results, utilizing adaptive threshold segmentation. This approach maximizes the strengths of foreground region detection and semantic segmentation algorithms, yielding superior results in combustible gas leakage foreground detection. Experimental results demonstrate the method’s enhanced accuracy in detecting combustible gas leakage under various conditions compared to existing foreground detection algorithms.
Infrared image recognition technology has a wide range of applications in the field of gas detection. Unlike visible light images, gas detection in infrared images is relatively difficult due to the lack of clear contrast and the relative blurriness of gas targets. This paper proposes a weakly supervised distillation network to address the issue of low detection accuracy of gas regions in infrared images in complex scenes. This method mainly generates accurate heatmaps as pseudo labels by utilizing complex class activation mappings; Using pseudo labels to train the specialized model proposed in this article, more accurate heat map results are generated, and finally the heat map results are fused with the foreground obtained based on background difference method to reduce false positives in combustible gas detection results. The experimental results show that the proposed method has high accuracy in various scenarios, and the model can efficiently run in embedded systems, effectively solving the problem of infrared gas recognition in complex scenarios.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.