TheiaNet: Towards fast and inexpensive CNN design choices for image dehazing

Mehra, Aryan; Narang, Pratik; Mandal, Murari

doi:10.1016/j.jvcir.2021.103137

Cited by 15 publications

(8 citation statements)

References 38 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…From Table 6, the proposed image-dehazing CNN leads to better results than the prior-based methods and learning-based methods that have been proposed in recent years. Compared to the latest approach, TheiaNet [20], the average SSIM of the indoor dehazing results increased by 5.65%, and the average PSNR increased by 7.73%, whereas the average SSIM of the outdoor dehazing results increased by 2.83%, and the average PSNR increased by 14.09%. [20], the average SSIM of the dehazing results increased by 2.47%, and the average PSNR increased by 8.09%.…”

Section: Performance and Discussionmentioning

confidence: 80%

“…When using this architecture to solve the problem of image dehazing, it is not possible to simply deepen the number of network layers; but a suitable neural network block should be imported. For example, TheiaNet [20] imports the bottleneck enhancer into the final output of the encoder, which extracts the feature map from coarse to fine through multi-scale pooling to obtain different feature maps and then concatenates these feature maps together. The final output of the decoder is added to the aggregation head, and the outputs of the encoder and decoder of different layers are upsampled to the same resolution and concatenated.…”

Section: Deep-learning-based Methodsmentioning

confidence: 99%

“…Tufail et al [23] experimentally proved that better contrast and brightness can be obtained in the YCbCr color space. Therefore, inspired by [22,23], Mehra et al [20] used a multi-cue color space, including RGB, HSV, YCbCr, and Lab, to formulate a 12-channel input.…”

Section: Multiple Input Channelmentioning

confidence: 99%

“…The proposed method integrates the concepts of GFN [21] and TheiaNet [20] to preprocess hazy images. In the next section, we introduce significant blocks of our imagedehazing CNN, including the image pre-processing blocks, inception-like blocks, SPP blocks, and attention gates (AG).…”

Section: Multiple Input Channelmentioning

confidence: 99%

“…DehazeNet [9] 8.305 K 0.035 MB AOD-Net [10] 1.76 K 0.008 MB ReViewNet [28] 399.7 K 5 MB FFA-Net [12] 4455.9 K 21.3 MB TheiaNet [20] 157.9 K 1.98 MB Proposed Method 207.3 K 0.86 MB…”

Section: Trainable Parameters Model Sizementioning

confidence: 99%

See 4 more Smart Citations

Attention-Gate-Based Model with Inception-like Block for Single-Image Dehazing

Tsai

Chen

2022

Applied Sciences

View full text Add to dashboard Cite

In recent decades, haze has become an environmental issue due to its effects on human health. It also reduces visibility and degrades the performance of computer vision algorithms in autonomous driving applications, which may jeopardize car driving safety. Therefore, it is extremely important to instantly remove the haze effect on an image. The purpose of this study is to leverage useful modules to achieve a lightweight and real-time image-dehazing model. Based on the U-Net architecture, this study integrates four modules, including an image pre-processing block, inception-like blocks, spatial pyramid pooling blocks, and attention gates. The original attention gate was revised to fit the field of image dehazing and consider different color spaces to retain the advantages of each color space. Furthermore, using an ablation study and a quantitative evaluation, the advantages of using these modules were illustrated. Through existing indoor and outdoor test datasets, the proposed method shows outstanding dehazing quality and an efficient execution time compared to other state-of-the-art methods. This study demonstrates that the proposed model can improve dehazing quality, keep the model lightweight, and obtain pleasing dehazing results. A comparison to existing methods using the RESIDE SOTS dataset revealed that the proposed model improves the SSIM and PSNR metrics by at least 5–10%.

show abstract

Section: Performance and Discussionmentioning

confidence: 80%

Section: Deep-learning-based Methodsmentioning

confidence: 99%

Section: Multiple Input Channelmentioning

confidence: 99%

Section: Multiple Input Channelmentioning

confidence: 99%

“…DehazeNet [9] 8.305 K 0.035 MB AOD-Net [10] 1.76 K 0.008 MB ReViewNet [28] 399.7 K 5 MB FFA-Net [12] 4455.9 K 21.3 MB TheiaNet [20] 157.9 K 1.98 MB Proposed Method 207.3 K 0.86 MB…”

Section: Trainable Parameters Model Sizementioning

confidence: 99%

See 3 more Smart Citations