diff --git a/tests/test_caption_text.py b/tests/test_caption_text.py new file mode 100644 index 0000000..082840f --- /dev/null +++ b/tests/test_caption_text.py @@ -0,0 +1,61 @@ +from __future__ import annotations + +import unittest +from unittest.mock import Mock + +from scene_search.caption_text import remove_image_preface +from scene_search.captioner import OllamaCaptioner + + +class CaptionTextTests(unittest.TestCase): + def test_removes_model_image_set_introductions(self) -> None: + examples = { + "The video features a series of still images showing a man wearing a red coat.": + "A man wearing a red coat.", + "A series of images depicts a woman standing beside a blue car.": + "A woman standing beside a blue car.", + "These frames show two people walking along a beach.": + "Two people walking along a beach.", + "This clip contains a collection of photographs depicting an old bridge at night.": + "An old bridge at night.", + "The images depict a small dog lying on the sofa.": + "A small dog lying on the sofa.", + } + for original, expected in examples.items(): + with self.subTest(original=original): + self.assertEqual(remove_image_preface(original), expected) + + def test_only_anchored_presentation_boilerplate_is_removed(self) -> None: + examples = [ + "A photographer is sorting a series of photos on a desk.", + "The video features a series of still images showing a", + "A row of framed pictures hangs behind a woman.", + "A person displays a sequence of images on a monitor.", + "A woman walks through a building and waves to the camera.", + ] + for original in examples: + with self.subTest(original=original): + self.assertEqual(remove_image_preface(original), original) + + def test_live_caption_generation_normalises_response_and_changes_prompt(self) -> None: + captioner = OllamaCaptioner("http://ollama.test", "vision-test") + response = Mock() + response.ok = True + response.json.return_value = { + "message": { + "content": ( + "The video features a series of still images showing a " + "person walking across a street." + ) + } + } + captioner.session.post = Mock(return_value=response) + result = captioner._caption_b64("an-image", chronological_sequence=True) + self.assertEqual(result, "A person walking across a street.") + prompt = captioner.session.post.call_args.kwargs["json"]["messages"][0]["content"] + self.assertIn("Never introduce your answer", prompt) + self.assertIn("Do not invent motion", prompt) + + +if __name__ == "__main__": + unittest.main()