Implement brightspot detection #228
Conversation
| # Object detections | ||
| with open(filename + ".txt", "w", encoding="utf-8") as file: | ||
| # Use internal string representation | ||
| file.write(repr(detections)) |
There was a problem hiding this comment.
Don't log files anymore, look at the new DetectTargetUltralytics for the desired logging (You will need to insert a parameter for passing in a logger)
| gray_image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY) | ||
|
|
||
| # Apply thresholding to isolate bright spots | ||
| _, bw_image = cv2.threshold(gray_image, BRIGHTSPOT_THRESHOLD, 255, cv2.THRESH_BINARY) |
There was a problem hiding this comment.
What is the first parameter? Why is it unused? If it tells apart success vs failure, please use an if or try/except and log error
| Return: Success, detections. | ||
| """ | ||
| image = data.image | ||
| gray_image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY) |
There was a problem hiding this comment.
put in try/except and logger.error any errors
| from .. import detections_and_time | ||
|
|
||
|
|
||
| BRIGHTSPOT_THRESHOLD = 240 |
There was a problem hiding this comment.
Could you make this a parameter that's passed in from config.yaml? It may be easier to test that way. For your unit tests you can continue to use a hard coded value. cc @Xierumeng do we want to modify detect target factory to have a dict (kind of like kwargs?) for these kinds of things?
There was a problem hiding this comment.
I dislike kwargs because it removes information from a function signature. Perhaps a data exchange format for the configs? For example:
detect_target_ultralytics.py
class DetectTargetUltralyticsConfig:
def __init__(self, thing, other_thing):
...
class DetectTargetUltralytics:
@classmethod
def create(cls, common_thing, another_common, config: DetectTargetUltralyticsConfig):
...detect_target_brightspot.py
class DetectTargetBrightspotConfig:
def __init__(self, other_thing, setting):
...
class DetectTargetBrightspot:
@classmethod
def create(cls, common_thing, another_common, config: DetectTargetBrightspotConfig):
...detect_target_factory.py
def create_detect_target(detect_target_option: DetectTargetOption, common_thing, another_common, config: DetectTargetUltralyticsConfig | DetectTargetBrightspotConfig):
match detect_target_option:
case DetectTargetOption.ML_ULTRALYTICS:
return detect_target_ultralytics.DetectTargetUltralytics.create(common_thing, another_common, config)
case DetectTargetOption.C_BRIGHTSPOT:
return detect_target_brightspot.DetectTargetBrightspot.create(common_thing, another_common, config)
return False, None|
|
||
| TEST_PATH = pathlib.Path("tests", "brightspot_example") | ||
|
|
||
| IMAGE_IR1_PATH = pathlib.Path(TEST_PATH, "ir.png") |
There was a problem hiding this comment.
Maybe rename your photo to ir1.png?
| size = keypoint.size | ||
| bounds = np.array([x - size / 2, y - size / 2, x + size / 2, y + size / 2]) | ||
| confidence = keypoint.response | ||
| result, detection = detections_and_time.Detection.create(bounds, 1, confidence) |
There was a problem hiding this comment.
Rather than using magic number 1, set as enum or a global constant at the top. Also, do you know what the keypoint.response is? It seems to be 0, which usually means it is not confident?
| Test detection on multiple copies of the IR image. | ||
| """ | ||
| image_count = 4 | ||
| input_images = [copy.deepcopy(image_ir) for _ in range(image_count)] |
There was a problem hiding this comment.
You should test on some new images that we collected from flight test. See oneDrive, 11-23 and 11-22 (pick before/after tracks, around 100.png for ~e=250).
Also find a pic that has no IR beacon to make sure it detects nothing
Xierumeng
left a comment
Xierumeng
left a comment
There was a problem hiding this comment.
Reviewed functionality, tests to be reviewed.
siddhp1
force-pushed
the
detect-brightspots
branch
from
0fe3eac to
8f18e9c
Compare
* Implement brightspot detection * Fix detections * Add to factory method * Add multiple testing images * Separate tests into detections and no detections * Fix minor issues * Fix comments * Fix image naming * Add check for no detections * Fix constant ordering
No description provided.