[Guix packaging] cysmith/neural-style-tf #108

New issue

Open

opened 2026-08-14 13:17:28 +00:00 by htayj · 1 comment

htayj commented

2026-08-14 13:17:28 +00:00

(Migrated from github.com)

Candidate

Upstream canonical URL: https://github.com/cysmith/neural-style-tf
Source pinned commit/release when known: a2c374f9ee2938f0022e1e0b720f4eb28cf7d0a8 on master (default branch snapshot reviewed 2026-08-14).
Target concrete installed deliverable: Neural-style image/video transfer scripts and Python API
Primary category: command-line-tool
Tags: image-processing, ai-ml
Primary language normalized: Python
Build system: Custom Python scripts (neural_style.py, stylize_image.sh, stylize_video.sh; no packaging manifest)
SPDX expression: GPL-3.0
License status: confirmed-free
License evidence: LICENSE contains GPLv3 and README documents the TensorFlow neural-style implementation.
Difficulty: hard — The project targets obsolete TensorFlow/model-download workflows with no dependency manifest; pin model files and provide a CPU/offline execution path.
Workflow state: research
Existing Guix coverage: Checked 2026-08-14: GNU Guix, Nonguix, Guix Science, Guix HPC, Guix Past, Guix 'R Us, and RDE; no equivalent package with the same upstream origin was found.

Scope and blockers

Package the pinned upstream source as the stated deliverable, retaining upstream notices and making runtime services, credentials, downloaded assets, and optional integrations explicit. The project targets obsolete TensorFlow/model-download workflows with no dependency manifest; pin model files and provide a CPU/offline execution path.

Acceptance checks

guix lint -L. neural-style-tf passes with no new errors.
guix build -L. neural-style-tf succeeds from the pinned source with tests enabled where practical.
App-specific offline smoke: Run a tiny local image fixture through the script with model data pre-seeded, asserting output creation without network.

Imported from GitHub issue/PR. Originally posted by htayj on 2026-08-14T13:17:28Z.

## Candidate - Upstream canonical URL: https://github.com/cysmith/neural-style-tf - Source pinned commit/release when known: `a2c374f9ee2938f0022e1e0b720f4eb28cf7d0a8` on `master` (default branch snapshot reviewed 2026-08-14). - Target concrete installed deliverable: Neural-style image/video transfer scripts and Python API - Primary category: command-line-tool - Tags: image-processing, ai-ml - Primary language normalized: Python - Build system: Custom Python scripts (`neural_style.py`, `stylize_image.sh`, `stylize_video.sh`; no packaging manifest) - SPDX expression: GPL-3.0 - License status: confirmed-free - License evidence: `LICENSE` contains GPLv3 and README documents the TensorFlow neural-style implementation. - Difficulty: hard — The project targets obsolete TensorFlow/model-download workflows with no dependency manifest; pin model files and provide a CPU/offline execution path. - Workflow state: research - Existing Guix coverage: Checked 2026-08-14: GNU Guix, Nonguix, Guix Science, Guix HPC, Guix Past, Guix 'R Us, and RDE; no equivalent package with the same upstream origin was found. ## Scope and blockers Package the pinned upstream source as the stated deliverable, retaining upstream notices and making runtime services, credentials, downloaded assets, and optional integrations explicit. The project targets obsolete TensorFlow/model-download workflows with no dependency manifest; pin model files and provide a CPU/offline execution path. ## Acceptance checks - `guix lint -L. neural-style-tf` passes with no new errors. - `guix build -L. neural-style-tf` succeeds from the pinned source with tests enabled where practical. - App-specific offline smoke: Run a tiny local image fixture through the script with model data pre-seeded, asserting output creation without network. --- Imported from [GitHub issue/PR](https://github.com/htayj/guix-channel/issues/108). Originally posted by [htayj](https://github.com/htayj) on 2026-08-14T13:17:28Z.

tay commented

2026-09-21 07:16:37 +00:00

Owner

Goocastle recorded disposition: blocked.

Blocked by the reviewed host toolchain manifest, not by source legality or viability. The channel has no neural-style-tf duplicate. Authoritative upstream is the fixed Git revision a2c374f9ee2938f0022e1e0b720f4eb28cf7d0a8 of https://github.com/cysmith/neural-style-tf; its LICENSE at that revision is GPLv3. It is a source-distributable Python 2-era TensorFlow 1 application importing tensorflow, numpy, scipy.io, and cv2, with a documented /cpu:0 device and --model_weights input. The separately required MatConvNet model is fixed by authoritative VLFeat metadata to https://www.vlfeat.org/matconvnet/models/imagenet-vgg-verydeep-19.mat at ETag 1fe1ffcf-53db4a482dac0 / Last-Modified Fri, 30 Sep 2016 07:35:47 GMT, size 534904783, MD5 106118b7cf60435e6d8e04f6a6dc3657; VGG's authoritative page releases VGG-VD models under CC BY 4.0. An image-only CPU package should therefore install neural_style.py with an upstream-license wrapper, synthesize a safe deterministic 1x1 RGB content image, a 1x1 RGB style image, and a minimal valid fixed VGG19 .mat fixture, then run neural_style.py in /cpu:0 mode with --optimizer adam --max_iterations 1 --max_size 1 --verbose and assert creation of the output PNG. It should omit video mode and the tracked x86-64 static optical-flow binaries. The approved reviewed manifest provides a Guix executable but no TF1-compatible provider: host python3 lacks tensorflow and cv2, and using guix shell is forbidden by this phase. An unattended implementation cannot be declared safe without a reviewed Guix package set containing tensorflow (1.x compatible with neural-style-tf), python-numpy, python-scipy, python-opencv, and optionally ffmpeg plus a MAT writer for the runtime fixture. Extend the approved manifest under review with those exact providers, or provide the selected TF1-compatible package derivations for authorization; then re-run this research/delivery gate and make daemon-backed build, lint, and the CPU offline PNG smoke required acceptance.

Goocastle recorded disposition: blocked. Blocked by the reviewed host toolchain manifest, not by source legality or viability. The channel has no neural-style-tf duplicate. Authoritative upstream is the fixed Git revision a2c374f9ee2938f0022e1e0b720f4eb28cf7d0a8 of https://github.com/cysmith/neural-style-tf; its LICENSE at that revision is GPLv3. It is a source-distributable Python 2-era TensorFlow 1 application importing tensorflow, numpy, scipy.io, and cv2, with a documented /cpu:0 device and --model_weights input. The separately required MatConvNet model is fixed by authoritative VLFeat metadata to https://www.vlfeat.org/matconvnet/models/imagenet-vgg-verydeep-19.mat at ETag 1fe1ffcf-53db4a482dac0 / Last-Modified Fri, 30 Sep 2016 07:35:47 GMT, size 534904783, MD5 106118b7cf60435e6d8e04f6a6dc3657; VGG's authoritative page releases VGG-VD models under CC BY 4.0. An image-only CPU package should therefore install neural_style.py with an upstream-license wrapper, synthesize a safe deterministic 1x1 RGB content image, a 1x1 RGB style image, and a minimal valid fixed VGG19 .mat fixture, then run neural_style.py in /cpu:0 mode with --optimizer adam --max_iterations 1 --max_size 1 --verbose and assert creation of the output PNG. It should omit video mode and the tracked x86-64 static optical-flow binaries. The approved reviewed manifest provides a Guix executable but no TF1-compatible provider: host python3 lacks tensorflow and cv2, and using guix shell is forbidden by this phase. An unattended implementation cannot be declared safe without a reviewed Guix package set containing tensorflow (1.x compatible with neural-style-tf), python-numpy, python-scipy, python-opencv, and optionally ffmpeg plus a MAT writer for the runtime fixture. Extend the approved manifest under review with those exact providers, or provide the selected TF1-compatible package derivations for authorization; then re-run this research/delivery gate and make daemon-backed build, lint, and the CPU offline PNG smoke required acceptance.

tay added the

state:blocked

label

2026-09-21 07:16:37 +00:00