TL;DR
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
Nibnib is an experimental image-compression format designed around fast, simple decoding, with a target use case of displaying images on an RP2350 microcontroller. Its author says the decoder processes two RGB 565 pixels from each encoded byte, while the current encoder is much slower and image quality could improve.
Nibnib, an experimental image-compression format described in a GitHub report, is designed to decode images with very little work per pixel on an RP2350 microcontroller. Its author says the decoder processes two pixels from each encoded byte using a palette of color deltas, targeting a tight display budget rather than archival storage or general-purpose image distribution.
The project began as an experiment in image quality and performance for DVI output on an RP2350, according to its report. The author describes a target budget of about six processor cycles per pixel on devices of this class. Nibnib uses a palette, but its entries represent color changes rather than complete colors: each selected delta is added to the previous pixel value, which is duplicated to form the starting point for the next pair.
In the described decoding process, the first two pixels of a row are stored directly. For later pairs, the decoder reads one byte, uses it to select one of 256 palette entries, and adds that 32-bit palette value to the duplicated previous pixel. The resulting low and high halves supply two RGB 565 pixels. This yields four bits per pixel for the encoded pairs, excluding the row’s initial pixels and format overhead. The author presents the method as deterministic and straightforward to decode; a decoder for the RP2350 is included in the image demo in the vic-2350 repository.
The encoder does substantially more work than the decoder. The report says its current implementation searches palette choices and codewords, choosing entries to reduce error and distributing that error with Floyd–Steinberg error diffusion. The palette combines a small cube of delta values with randomly selected values. The author says the small-value cube helps avoid artifacts near dark and bright limits. The palette is not currently adapted to each input image, and the report identifies that as one potential route to better image quality.
Speed for Microcontroller Displays
Nibnib addresses a specific engineering trade-off: a device with limited processing time may need an image representation it can decode as it draws, rather than spending more resources on a more complex decompression routine. Processing two pixels per input byte and reusing the previous pixel aims to keep that work small enough for the project’s stated display target. If the method proves useful in practice, it could make compressed images more practical in tightly constrained graphics pipelines.
The report does not establish that Nibnib outperforms established formats across devices, image types, or quality settings. Its value is currently as a working experiment with a narrowly defined use case. The trade-off is also asymmetric: decoding is intended to be cheap, but encoding is currently much slower. That may be acceptable when images are prepared in advance and displayed repeatedly, but the source does not provide timings or comparisons to quantify the benefit.
Products Worth Considering
Amazon Kindle Scribe Colorsoft, 64GB, 11-Inch Color Display with Premium Pen
Bose QuietComfort True Wireless Earbuds with Active Noise Cancellation, Black
Bose SoundLink Plus Portable Bluetooth Speaker, Black
As an affiliate, we earn on qualifying purchases.
A Format Defined by Its Decoder
The report describes Nibnib as a format defined by its decoding procedure. The decoder’s work is fixed and simple, while the encoder can search in different ways for byte sequences that reproduce an image with acceptable error. This makes the current format compatible with possible improvements to encoding, including a more carefully structured or image-specific palette and a deeper search for pixel choices.
The file layout is deliberately basic: two bytes each for width and height, followed by 256 32-bit palette entries and image rows. Each row starts with two RGB 565 pixels, then uses palette-index bytes for subsequent pairs; row stride is rounded to a multiple of four bytes. The report says the format has no other metadata and is not intended for archival use. It is more likely to be integrated closely with a decoder. The project is licensed under either Apache 2.0 or MIT, according to the supplied report.
“The image format is tuned for extreme decompression speed.”
— Nibnib project report on GitHub
Quality and Performance Still Open
The report provides no benchmark results showing actual cycles per pixel, file sizes, or decoding speed relative to other image formats. The six-cycle figure is presented as the target budget for the device class, not as a measured Nibnib result. It also does not provide a standardized image-quality evaluation or specify how results vary across different kinds of images.
Image quality remains an active limitation in the report’s account. The palette is not tailored to the source image, and the greedy encoding process can leave artifacts trailing to the right of sharp edges. The author says improvements to palette selection and a deeper search could improve results, but they would increase encoder work. It is not clear whether those changes have been implemented, how the format handles dimensions outside the described row scheme, or whether the project will be developed into a broader-use format.
Potential Encoder Improvements
The report points to several possible next steps: optimize palette selection for each image, search more deeply for codewords that improve later encoding choices, and speed up the current brute-force palette search. It also describes encoding from right to left or alternating direction by row as a way to address edge artifacts, while retaining left-to-right decoding. These are proposed directions, not reported completed features.
Further project updates, source-code changes, or testing on the target hardware would clarify whether the approach meets its performance and quality goals. The supplied report gives no schedule for those developments, so the next milestone and the format’s wider adoption remain unknown.
Key Questions
What is Nibnib?
Nibnib is an experimental image format designed to make decompression simple and fast, with an initial focus on displaying images through DVI output on an RP2350 microcontroller.
How does its decoder produce two pixels from one byte?
After the first two pixels in a row, each byte indexes a 256-entry palette. The selected color delta is added to the previous pixel, duplicated across a pair, to generate two RGB 565 pixels.
Has Nibnib been shown to meet its speed target?
The report describes a target budget of about six cycles per pixel, but the supplied material gives no measured benchmark confirming that Nibnib meets it or comparing it with other formats.
Is Nibnib meant for archival images?
No. The project report says the format is not intended for archival and is more likely to be used in close integration with its decoder.
What improvements are being considered?
The report identifies image-specific palette selection, deeper searches during encoding, a faster palette search, and changes to encoding direction as possible improvements. It does not say these changes have been completed or give a release timeline.
Source: github.com
