Find files, editable templates and browser test targets by what you need to make or test. The directory below is cut by format; the two collections under it cut the same library by subject and by workflow.
A depth map for the published plate d-menu.png, 1497x1024. Monocular depth from Depth Anything V2 (vitl), rendered at the plate's long edge rather than the 512-pixel default, so depth and colour can be compared per pixel without a resample in between. Lit coverage measures 74.6% of the frame. The plate and this map are the same scene at the same size, so they can be compared pixel for pixel rather than by eye.
The same view of a restaurant menu card standing on a laid table, cutlery and glassware, warm evening light, expanded to 704x704 - 96 pixels added on all four sides - so 47.1% of this frame is invented canvas. The original sits at pixels 96,96 to 608,608. Compare it INSET by 40 pixels: over that core it differs from the original by only 7.121/255, the VAE round trip, but over the whole rectangle by 12.793/255, because the pad deliberately feathers the original's outer edge into the new area. A test that expects the whole rectangle untouched fails on a correct tool.
A 512x512 view of a restaurant menu card standing on a laid table, cutlery and glassware, warm evening light - the frame an outpainting tool is given, cropped from the published plate nss-d-menu_00001_.png. Its companion expands it to 704x704 with 96 pixels added on all four sides, and the boundary record in this group states the exact rectangle this image occupies inside that frame - so where it ended up can be checked rather than eyeballed.
A tangent-space surface normal map for the published plate d-menu.png, 1216x832 - the same size as the plate, so the two compare pixel for pixel with no resample in between. RGB encodes the XYZ surface direction remapped from -1..1 into 0..255. Decoded back to vectors this file measures a mean length of 0.9963, which is the number to check your own decode against: a reader that transposes the channels or inverts the remap still produces a plausible-looking image, and lights the surface the wrong way.
A canny edge map for the published plate d-receipts.png, 1216x832. Canny edge detection at low threshold 0.1 and high 0.3, run at the plate's own resolution so the edges land on the same pixels as the photograph they came from. Lit coverage measures 11.5% of the frame. The plate and this map are the same scene at the same size, so they can be compared pixel for pixel rather than by eye.
A depth map for the published plate d-receipts.png, 1497x1024. Monocular depth from Depth Anything V2 (vitl), rendered at the plate's long edge rather than the 512-pixel default, so depth and colour can be compared per pixel without a resample in between. Lit coverage measures 97.7% of the frame. The plate and this map are the same scene at the same size, so they can be compared pixel for pixel rather than by eye.
The same view of a scatter of crumpled paper receipts on a kitchen table, overhead view, harsh afternoon sun and hard shadows, expanded to 768x640 - a different amount on every side, which is the case a tool assuming symmetry fails - so 46.7% of this frame is invented canvas. The original sits at pixels 64,32 to 576,544. Compare it INSET by 40 pixels: over that core it differs from the original by only 8.885/255, the VAE round trip, but over the whole rectangle by 15.578/255, because the pad deliberately feathers the original's outer edge into the new area. A test that expects the whole rectangle untouched fails on a correct tool.
A 512x512 view of a scatter of crumpled paper receipts on a kitchen table, overhead view, harsh afternoon sun and hard shadows - the frame an outpainting tool is given, cropped from the published plate nss-d-receipts_00001_.png. Its companion expands it to 768x640 with a different amount on every side, which is the case a tool assuming symmetry fails, and the boundary record in this group states the exact rectangle this image occupies inside that frame - so where it ended up can be checked rather than eyeballed.
A tangent-space surface normal map for the published plate d-receipts.png, 1216x832 - the same size as the plate, so the two compare pixel for pixel with no resample in between. RGB encodes the XYZ surface direction remapped from -1..1 into 0..255. Decoded back to vectors this file measures a mean length of 0.9953, which is the number to check your own decode against: a reader that transposes the channels or inverts the remap still produces a plausible-looking image, and lights the surface the wrong way.
The same plate encoded once at JPEG quality 25. Blocking and ringing are bounded by that single pass, so a restorer can be scored against the PNG reference in this group rather than against an opinion Plate g-dungeon.
The same plate encoded once at JPEG quality 50. Blocking and ringing are bounded by that single pass, so a restorer can be scored against the PNG reference in this group rather than against an opinion Plate g-dungeon.
The same plate encoded once at JPEG quality 75. Blocking and ringing are bounded by that single pass, so a restorer can be scored against the PNG reference in this group rather than against an opinion Plate g-dungeon.
The same plate encoded once at JPEG quality 90. Blocking and ringing are bounded by that single pass, so a restorer can be scored against the PNG reference in this group rather than against an opinion Plate g-dungeon.
A synthetic photographic plate of a stylised rendered game scene or character, 1024x1024, lossless PNG. This is the reference every lossy variant in this group is derived from Plate g-dungeon.
The same plate encoded once at JPEG quality 25. Blocking and ringing are bounded by that single pass, so a restorer can be scored against the PNG reference in this group rather than against an opinion Plate t-ph-dark1.
The same plate encoded once at JPEG quality 50. Blocking and ringing are bounded by that single pass, so a restorer can be scored against the PNG reference in this group rather than against an opinion Plate t-ph-dark1.
The same plate encoded once at JPEG quality 75. Blocking and ringing are bounded by that single pass, so a restorer can be scored against the PNG reference in this group rather than against an opinion Plate t-ph-dark1.
The same plate encoded once at JPEG quality 90. Blocking and ringing are bounded by that single pass, so a restorer can be scored against the PNG reference in this group rather than against an opinion Plate t-ph-dark1.
A synthetic photographic plate of a soft low-detail background plate, 1024x683, lossless PNG. This is the reference every lossy variant in this group is derived from Plate t-ph-dark1.
The same plate encoded once at JPEG quality 25. Blocking and ringing are bounded by that single pass, so a restorer can be scored against the PNG reference in this group rather than against an opinion Plate t-ph-dark2.
The same plate encoded once at JPEG quality 50. Blocking and ringing are bounded by that single pass, so a restorer can be scored against the PNG reference in this group rather than against an opinion Plate t-ph-dark2.
The same plate encoded once at JPEG quality 75. Blocking and ringing are bounded by that single pass, so a restorer can be scored against the PNG reference in this group rather than against an opinion Plate t-ph-dark2.
