Ambiq/compressionkit-e2a2-image-v0.3

E2A2 image codec for compressionkit

E2A2 (Encoding Efficient Asymmetric Autoencoders) image is an image codec for 8-bit RGB images; rates are bits per pixel. This repository holds its weights in the layout that compressionkit v0.3 reads, with an encoder half and a decoder half that load separately.

Source and permission

The weights are redistributed from their original public release, danjacobellis/E2A2, subfolder image/, revision 2c4f517c, with the permission of the author, Dan Jacobellis. Ambiq did not train this codec. Its description, development and evaluation are documented at the original source and on the codec's page, https://danjacobellis.net/compressors/E2A2_image. The entropy coder is GGDLPC (https://danjacobellis.net/GGDLPC/).

Every tensor here equals the tensor of the same name at that revision; this repository divides them into the encoder and decoder halves listed below and adds the files compressionkit uses.

Files

• encoder/config.json, encoder/weights.safetensors (the merged.{op}.encoders.* tensors) and encoder/entropy.json (the GGDLPC tables): everything needed to encode.
• decoder/config.json, decoder/weights.safetensors (the merged.{op}.decoder.* tensors): everything needed to decode. Each config.json is the full source configuration, so each half constructs alone.
• rd.json: the rate–distortion curve of the original release, byte-identical to it. The curve is on the 24 Kodak images, bits per pixel, PSNR on the RGB image mapped to [0, 1]. Nothing in the codec or its entropy coder was trained or fitted on Kodak.
• web/: the decoder of each operating point that the compressionkit browser demo offers, as LiteRT files (section below). The compressionkit loaders do not download web/.

file	bytes	SHA-256
decoder/config.json	1,705	6903c5761232f47cc403c346f3de62cf146479dc5e3fa1466ad1c060731a9ad8
decoder/weights.safetensors	2,129,494,748	7b67591da5286f595d1cebb0ddc3fefb854df60c18ae6181369e57425c1faa5b
encoder/config.json	1,705	6903c5761232f47cc403c346f3de62cf146479dc5e3fa1466ad1c060731a9ad8
encoder/entropy.json	8,617	a2f231e659315bcd6de5b227252fb2d32fd62b0ebb54f95672996454ac87e7d2
encoder/weights.safetensors	280,812	ad615d04ea5ef07de2e3133cf6a05ac54329ef3261990db664561b324c32c820
rd.json	865	c9aad3b3ee342712b3b8df36e0bd4a398f4ca96e964820fecda993d5b86f73f0

Input, output and operating points

Input: float32 (3, H, W) = 8-bit sRGB / 127.5 − 1, cropped to whole 64-pixel macroregions. Output: a reconstruction in the same form. Operating points (cumulative latent channels): [3, 9, 15, 21, 27]. The stream of a lower operating point is a literal bit-prefix of the stream of the highest, and the rates are entropy-coded bits. compressionkit converts nothing: each codec states its input type and range.

Using it with compressionkit

compressionkit v0.3 is in pre-release; these loaders are part of it.

from compressionkit.e2a2 import hub

# both halves
config, weights, coder, ops = hub.load_split_from_hub('image')
model = hub.load_codec(config=config, weights=weights, n_ch=ops[-1])

# one half (encode-only shown; half='decoder' for decode-only)
cfg, w, coder, ops = hub.load_split_from_hub('image', half='encoder')
encoder = hub.build_partial(cfg, w, ops[-1])

Status and limitations

• A reference codec for evaluation and development with compressionkit, redistributed as released; not a production model, and not qualified for any product or regulated use.
• Its quality on data unlike the data of its original release is not characterized here; measure it on your own data (compressionkit.calibrate fits an input map on your training split).
• Maintainer for compressionkit: Dan Jacobellis (Ambiq). Questions about the codec itself: the original source.

Browser decoders (web/)

For each operating point the compressionkit browser demo offers, its decoder as a LiteRT .tflite window decoder, which runs in a web browser on LiteRT.js (CPU), with a .json beside it that gives its shapes and windowing. A window decodes chunk output samples (pixels along each side for the image codec) with margin of context on each side; the latent grid is cut into overlapping windows and the outputs joined. The PSNR column is the largest difference between the PSNR of the joined output, decoded in LiteRT.js in Chromium, and that of the full-size PyTorch decode, over the example recordings of the demo. The compressionkit loaders do not download web/.

file	operating point	chunk	margin	bytes	PSNR difference (dB, rows)	decoder weights (SHA-256)
web/image_op3_chunk256_m64.tflite	3	256 pixels	64	228,033,684	1.9e-03 (1)	7b67591da528
web/image_op9_chunk256_m64.tflite	9	256 pixels	64	228,199,572	1.2e-04 (1)	7b67591da528
web/image_op15_chunk256_m64.tflite	15	256 pixels	64	455,569,908	1.0e-05 (1)	7b67591da528
web/image_op21_chunk256_m64.tflite	21	256 pixels	64	607,703,092	1.9e-04 (1)	7b67591da528
web/image_op27_chunk256_m64.tflite	27	256 pixels	64	610,357,300	1.3e-04 (1)	7b67591da528

Weights fp32, converted from decoder/weights.safetensors with litert-torch 0.10.0. The last column gives the start of the SHA-256 of the decoder/weights.safetensors it was converted from, which is this repository's.