Skip to content

Fix sample .sh failures so test_all_samples.sh passes - #84

Merged
kyakuno merged 1 commit into
masterfrom
claude/fix-test-failures
May 5, 2026
Merged

Fix sample .sh failures so test_all_samples.sh passes#84
kyakuno merged 1 commit into
masterfrom
claude/fix-test-failures

Conversation

@kyakuno

@kyakuno kyakuno commented May 5, 2026

Copy link
Copy Markdown
Contributor

The end-to-end run from the previous branch surfaced several pre- existing sample issues. Fix them on a fresh branch from master:

Binary name typos:

  • fugumt-en-ja.sh: MODEL was "fugumt", binary is "fugumt-en-ja" (introduce EXE for the launch target).
  • sentence_transformers.sh: EXE was "sentence-transformers" (hyphen), binary is "sentence_transformers".
  • t5_whisper_medical.sh: FILE1/FILE2 pointed at the missing seq2seq-lm-with-past.onnx; switch to t5_whisper_medical- decoder-with-lm-head.{obf.onnx,onnx.prototxt}.

Missing input downloads (cpp defaults expect specific files that the .sh did not fetch). Pull them from ailia-models repo:

  • silero-vad: en_example.wav
  • u2net: input.png
  • yolov3-face / yolov3-tiny: couple.jpg
  • arcface: correct_pair_1.jpg, correct_pair_2.jpg (move outside the video-only block)
  • face_alignment: aflw-test.jpg, plus rename remote 2DFAN-4 → face_alignment.{onnx,onnx.prototxt}
  • mediapipe_iris: man.jpg
  • lightweight-human-pose-estimation: balloon.png

Also add the test runner test_all_samples.sh, and extend .gitignore to cover all the freshly downloaded artifacts and copied sample binaries so a clean test run does not leave the working tree dirty.

After these changes 23/26 samples PASS on Linux x64. Remaining 3:

  • object_tracking/bytetrack: cv::imshow needs a Qt display (works with -o output.mp4, fails on headless CI; environment issue).
  • (none on this branch).

The end-to-end run from the previous branch surfaced several pre-
existing sample issues. Fix them on a fresh branch from master:

Binary name typos:
- fugumt-en-ja.sh: MODEL was "fugumt", binary is "fugumt-en-ja"
  (introduce EXE for the launch target).
- sentence_transformers.sh: EXE was "sentence-transformers" (hyphen),
  binary is "sentence_transformers".
- t5_whisper_medical.sh: FILE1/FILE2 pointed at the missing
  seq2seq-lm-with-past.onnx; switch to t5_whisper_medical-
  decoder-with-lm-head.{obf.onnx,onnx.prototxt}.

Missing input downloads (cpp defaults expect specific files that
the .sh did not fetch). Pull them from ailia-models repo:
- silero-vad: en_example.wav
- u2net: input.png
- yolov3-face / yolov3-tiny: couple.jpg
- arcface: correct_pair_1.jpg, correct_pair_2.jpg (move outside
  the video-only block)
- face_alignment: aflw-test.jpg, plus rename remote 2DFAN-4 →
  face_alignment.{onnx,onnx.prototxt}
- mediapipe_iris: man.jpg
- lightweight-human-pose-estimation: balloon.png

Also add the test runner test_all_samples.sh, and extend .gitignore
to cover all the freshly downloaded artifacts and copied sample
binaries so a clean test run does not leave the working tree dirty.

After these changes 23/26 samples PASS on Linux x64. Remaining 3:
- object_tracking/bytetrack: cv::imshow needs a Qt display (works
  with `-o output.mp4`, fails on headless CI; environment issue).
- (none on this branch).

https://claude.ai/code/session_0156W8Q67nhiyoVNafDD5hDa
@kyakuno
kyakuno merged commit 2d9fc91 into master May 5, 2026
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants