Thumbnail Generation and Storage¶
Introduction¶
PhotoPrism can use libvips and the native disintegration/imaging package to generate thumbnails for display in search results and in the full-screen viewer.
The smallest configurable size limit is 720px, which corresponds to the fit_720 thumbnail size. Most sizes up to 720x720 will be generated by default and should therefore always be available.
Besides their obvious use in the frontend, the indexer also uses thumbnails for color detection, face recognition and image classification, see Advanced Settings in the User Guide.
Standard Sizes¶
The smallest configurable static and dynamic size limit is 720px, so most sizes up to fit_720 are always generated by default.
Higher size limits generate thumbnails with more detail at higher resolutions - either statically (pre-generated while indexing) or on demand if the configuration permits.
Optional thumbnail sizes cannot be pre-generated and are only rendered on request, for example when sharing an image on Instagram.
The following overview shows the name, dimensions, and aspect ratio for each thumbnail size as well as a description of how it is used:
| Name | Width | Height | Aspect Ratio | Available | Usage |
|---|---|---|---|---|---|
| colors | 3 | 3 | 1:1 | Always | Color Detection |
| tile_50 | 50 | 50 | 1:1 | Always | List View |
| tile_100 | 100 | 100 | 1:1 | Always | Places View |
| left_224 | 224 | 224 | 1:1 | On-Demand | TensorFlow |
| right_224 | 224 | 224 | 1:1 | On-Demand | TensorFlow |
| tile_224 | 224 | 224 | 1:1 | Always | TensorFlow, Mosaic View |
| tile_500 | 500 | 500 | 1:1 | Always | Cards View |
| fit_720 | 720 | 720 | Preserved | Always | SD TV, Mobile |
| tile_1080 | 1080 | 1080 | 1:1 | Optional | |
| fit_1280 | 1280 | 1024 | Preserved | On-Demand | HD TV, SXGA |
| fit_1600 | 1600 | 900 | Preserved | Optional | Social Media |
| fit_1920 | 1920 | 1200 | Preserved | On-Demand | Full HD |
| fit_2048 | 2048 | 2048 | Preserved | Optional | DCI 2K, Tablets |
| fit_2560 | 2560 | 1600 | Preserved | On-Demand | Quad HD, Notebooks |
| fit_3840 | 3840 | 2400 | Preserved | Optional | 4K Ultra HD |
| fit_4096 | 4096 | 4096 | Preserved | On-Demand | DCI 4K, Retina 4K |
| fit_7680 | 7680 | 4320 | Preserved | On-Demand | 8K Ultra HD 2 |
Color Profiles¶
sRGB is the standard color space used when generating thumbnails. An ICC color profile for wide-gamut displays can optionally be embedded.
File Storage¶
Generated thumbnail files are stored in the storage/cache/thumbnails folder, where the path and file name depend on the thumbnail size and original file hash, e.g.:
storage/cache/thumbnails/1/a/3/1a30c1f...9_100x100_center.jpg
Downscaling Filters¶
Linear¶
Bilinear interpolation takes a weighted average of the four neighborhood pixels to calculate its final interpolated value. The result is a much smoother image than the original image. When all known pixel distances are equal, then the interpolated value is simply their sum divided by four. This technique performs interpolation in both directions, horizontal and vertical. This technique gives better result than nearest neighbor interpolation and take less computation time compared to bicubic interpolation.
Cubic¶
Catmull-Rom is a local interpolating spline developed for computer graphics purposes. Its initial use was in design of curves and surfaces, and has recently been used in several applications. Catmull-Rom splines are a family of cubic interpolating splines formulated such that the tangent at each point is calculated using the previous and next point on the spline. The results are similar to ones produced by bicubic interpolation with regards to sharpness, but the Catmull-Rom reconstruction is clearly superior in smooth signal region.
Lanczos¶
The Lanczos interpolation function is a mathematical formula used to smoothly interpolate the value of a digital image between its samples. It maps each sample of the given image to a translated and scaled copy of the Lanczos kernel, which is a sinc function windowed by the central hump of a dilated sinc function. The sum of these translated and scaled kernels is then evaluated at the desired pixel. Lanczos interpolation has the best properties in terms of detail preservation and minimal generation of aliasing artifacts for geometric transformations not involving strong down sampling. However higher order Lanczos interpolation requires high computational time, which makes it unsuitable for most commercial software.
Blackman¶
Blackman is a modification of Lanczos that has better control of ringing artifacts.
Server API¶
Like most commercial image hosting services, we have chosen to implement a cookie-free thumbnail API to minimize request latency by avoiding unnecessary network traffic.