From dc9105cc677e81f4f11ee08a7ca3bfb5f7ce735b Mon Sep 17 00:00:00 2001 From: jrobble Date: Wed, 11 Sep 2024 22:18:13 -0400 Subject: [PATCH 1/3] Fix width/height bug. Perform reverse transform. --- python/LlavaDetection/llava_component/llava_component.py | 9 +++++---- 1 file changed, 5 insertions(+), 4 deletions(-) diff --git a/python/LlavaDetection/llava_component/llava_component.py b/python/LlavaDetection/llava_component/llava_component.py index 4792c896..4cdd5f55 100644 --- a/python/LlavaDetection/llava_component/llava_component.py +++ b/python/LlavaDetection/llava_component/llava_component.py @@ -80,7 +80,7 @@ def get_detections_from_video(self, video_job: mpf.VideoJob) -> Iterable[mpf.Vid return tracks - def _get_frame_detections(self, media, config, is_video_job=False): + def _get_frame_detections(self, reader, config, is_video_job=False): self._update_prompts(config.prompt_config_path) self._check_client(config.ollama_server) @@ -90,11 +90,11 @@ def _get_frame_detections(self, media, config, is_video_job=False): video_process_timer = Timer() video_decode_timer.start() - for idx, frame in enumerate(media): + for idx, frame in enumerate(reader): video_decode_timer.pause() frame_count += 1 - width, height, _ = frame.shape + height, width, _ = frame.shape detection_properties = dict() self._get_ollama_response(self.frame_prompts, frame, detection_properties, video_process_timer) @@ -105,7 +105,8 @@ def _get_frame_detections(self, media, config, is_video_job=False): else: tracks.append(img_location) - video_decode_timer.start() + for track in tracks: + reader.reverse_transform(track) if is_video_job: return tracks, video_process_timer, video_decode_timer, frame_count From cfcbe6a6e4214a987c4a1bf8d0900e2ebe4bd4ab Mon Sep 17 00:00:00 2001 From: jrobble Date: Wed, 11 Sep 2024 22:24:34 -0400 Subject: [PATCH 2/3] Use BUILD_REGISTRY var. --- python/LlavaDetection/Dockerfile | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/python/LlavaDetection/Dockerfile b/python/LlavaDetection/Dockerfile index 8b4e3b00..128ee267 100644 --- a/python/LlavaDetection/Dockerfile +++ b/python/LlavaDetection/Dockerfile @@ -28,7 +28,7 @@ ARG BUILD_REGISTRY ARG BUILD_TAG=latest -FROM openmpf/openmpf_python_executor_ssb:${BUILD_TAG} +FROM ${BUILD_REGISTRY}openmpf_python_executor_ssb:${BUILD_TAG} RUN --mount=type=tmpfs,target=/var/cache/apt \ --mount=type=tmpfs,target=/var/lib/apt/lists \ From 2a9cbf8bee44c01aea87fa4f67ecd4db6f29dd99 Mon Sep 17 00:00:00 2001 From: jrobble Date: Wed, 11 Sep 2024 23:03:14 -0400 Subject: [PATCH 3/3] Add video_decode_timer.start() back. --- python/LlavaDetection/llava_component/llava_component.py | 2 ++ 1 file changed, 2 insertions(+) diff --git a/python/LlavaDetection/llava_component/llava_component.py b/python/LlavaDetection/llava_component/llava_component.py index 4cdd5f55..434b9f23 100644 --- a/python/LlavaDetection/llava_component/llava_component.py +++ b/python/LlavaDetection/llava_component/llava_component.py @@ -105,6 +105,8 @@ def _get_frame_detections(self, reader, config, is_video_job=False): else: tracks.append(img_location) + video_decode_timer.start() + for track in tracks: reader.reverse_transform(track)