Neural Distributed Image Compression Using Common Information

Mital, Nitish; Ezgi, Ozyilkan,; Ali, Garjani,; Gündüz, Deniz

doi:10.1109/dcc52660.2022.00026

Cited by 13 publications

(24 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Unlike the SAM approach [35], used also in SASIC [33], the CAM technique we introduce computes the attention globally, between patches of the latent representations over all channels, similarly to [11]. We show that our method outperforms the solution provided in [22]. We also show that our method is able to perform well in the case of unsynchronized and uncalibrated stereo cameras, that is, when the correlated images are generated at different time steps.…”

Section: Distributed Stereo Compressionmentioning

confidence: 91%

“…In the current literature, the DNN-based methods that explicitly address distributed (stereo) image compression are: (1) the DRASIC model in [10], (2) the DSIN model [2], (3) the NDIC model in [22], and (4) vector quantized variational autoencoder-based approach in [32]. In [2], the authors exploit high spatial correlations between pairs of stereo images, having significantly overlapping fields of view.…”

Section: Distributed Stereo Compressionmentioning

confidence: 99%

Section: Distributed Stereo Compressionmentioning

confidence: 99%

“…The paper [22] uses a different approach by explicitly modeling the correlation between the two stereo images. More precisely, [22] models the two images as being generated by a common set of features, as well as two independent sets of features that capture the information in the respective images that is not captured by the set of common features. The encoder only sends the independent information corresponding to the input image, in order to minimize the redundant information that is transmitted, while the set of common features between the input image and the side information are recovered locally only from the latter.…”

Section: Distributed Stereo Compressionmentioning

confidence: 99%

“…In this paper, we use the NDIC model [22] as the "backbone", which itself is built upon the model in [3] as its backbone. In principle, any other single image compression algorithm can also be used as the backbone for NDIC; and hence, also for our model.…”

Section: Distributed Stereo Compressionmentioning

confidence: 99%

See 4 more Smart Citations

Neural Distributed Image Compression with Cross-Attention Feature Alignment

Mital¹,

Ezgi²,

Ali³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

We propose a novel deep neural network (DNN) architecture for compressing an image when a correlated image is available as side information only at the decoder side, a special case of the well-known and heavily studied distributed source coding (DSC) problem. In particular, we consider a pair of stereo images, which have overlapping fields of view, captured by a synchronized and calibrated pair of cameras; and therefore, are highly correlated. We assume that one image of the pair is to be compressed and transmitted, while the other image is available only at the decoder. In the proposed architecture, the encoder maps the input image to a latent space using a DNN, quantizes the latent representation, and compresses it losslessly using entropy coding. The proposed decoder extracts useful information common between the images solely from the available side information, as well as a latent representation of the side information. Then, the latent representations of the two images, one received from the encoder, the other extracted locally, along with the locally generated common information, are fed to the respective decoders of the two images. We employ a cross-attention module (CAM) to align the feature maps obtained in the intermediate layers of the respective decoders of the two images, thus allowing better utilization of the side information. We train and demonstrate the effectiveness of the proposed algorithm on various realistic setups, such as KITTI and Cityscape datasets of stereo image pairs. Our results show that the proposed architecture is capable of exploiting the decoder-only side information in a more efficient manner as it outperforms previous works. We also show that the proposed method is able to provide significant gains even in the case of uncalibrated and unsynchronized camera array use cases.

show abstract

Section: Distributed Stereo Compressionmentioning

confidence: 91%