scCO2-Brine-Glass Dataset for Comparing Image Denoising Algorithms

32-bit, 1426 images. Low-exposure low quality dataset. The coreholder and jacket have been cropped out. The top and bottom of the sample were cropped out to eliminate the cone-beam effect, which is typically observed at the top and bottom of the scan volume

Segmented

No

Voxel length (x, y, z)

15.0, 15.0, 15.0 um

Sample

View Sample: Glass (borosilicate) Core

Analysis of this Data

• (Traditional Filter) Non-Local Means - Denoised
• (Traditional Filter) Gaussian - Denoised
• (Traditional Filter) Median - Denoised
• (Traditional Filter) Anisotropic Diffusion - Denoised
• (Traditional Filter) Bilateral - Denoised
• (Traditional Filter) Symmetric Nearest Neighbor - Denoised
• (Supervised DL Model) Noise2Clean - Denoised
• (Supervised DL Model) Cycle Consistent Generative Adversarial Network - Denoised
• (Supervised DL Model) Residual Dense Network - Denoised
• (Unsupervised DL Model) Noise2Void - Denoised
• (Semi-supervised DL Model) Noise2Noise (50%) - Denoised
• (Unsupervised DL Model) Noise2Noise - Denoised
• (Semi-supervised DL Model) Noise2Noise (75%) - Denoised
• (Semi-supervised DL Model) Noise2Noise (25%) - Denoised
• LQ (Segmented)
• Non-Local Means (Segmented)
• Gaussian (Segmented)
• Median (Segmented)
• Anisotropic Diffusion (Segmented)
• Bilateral (Segmented)
• Symmetric Nearest Neighbor (Segmented)
• Noise2Clean (Segmented)
• Cycle Consistent Generative Adversarial Network (Segmented)
• Residual Dense Network (Segmented)
• Noise2Void (Segmented)
• Noise2Noise (50%) (Segmented)
• Noise2Noise (Segmented)
• Noise2Noise (75%) (Segmented)
• Noise2Noise (25%) (Segmented)