OC-SORT on top of 2.1.0 #207
Conversation
| minimum_consecutive_frames: int = 3, # should change this for min_hits? | ||
| minimum_iou_threshold: float = 0.3, | ||
| direction_consistency_weight: float = 0.2, | ||
| high_conf_det_threshold: float = 0.6, |
There was a problem hiding this comment.
I want argument names to be easy to understand. Avoid abbreviations. Keep them short and clear.
There was a problem hiding this comment.
| high_conf_det_threshold: float = 0.6, | |
| high_conf_detect_threshold: float = 0.6, |
There was a problem hiding this comment.
i can change it here, but we need to also change it in bytetrack @SkalskiP
trackers/core/ocsort/tracker.py
Outdated
| self, | ||
| lost_track_buffer: int = 30, | ||
| frame_rate: float = 30.0, | ||
| minimum_consecutive_frames: int = 3, # should change this for min_hits? |
There was a problem hiding this comment.
re: should change this for min_hits? why?
There was a problem hiding this comment.
it was a now removed comment, because in the original version there was that parameter, but in our api it translates to that
trackers/core/ocsort/tracker.py
Outdated
| direction_consistency_matrix: Direction of the tracklet consistency cost matrix. | ||
|
|
||
| Returns: | ||
| tuple[list[tuple[int, int]], list[int], list[int]]: Matched indices, |
There was a problem hiding this comment.
Let's drop types here as well.
There was a problem hiding this comment.
Do it for all the classes/methods/functions we have in our API.
| @@ -0,0 +1,261 @@ | |||
| import numpy as np | |||
There was a problem hiding this comment.
Take a look at current develop. I added license headers to all our files. Make sure to include them as well.
| import numpy as np | ||
|
|
||
|
|
||
| def xyxy_to_xcycsr(xyxy: np.ndarray) -> np.ndarray: |
There was a problem hiding this comment.
Nice. Make sure we add those unit tests.
There was a problem hiding this comment.
where do unit tests go?
| return np.array([x, y, s, r]) | ||
|
|
||
|
|
||
| def xcycsr_to_xyxy(xcycsr: np.ndarray) -> np.ndarray: |
There was a problem hiding this comment.
I'm pretty sure we could make it faster with inplace operations. I jsut vibe-coded this version, but please dig a bit deeper into this.
def xcycsr_to_xyxy_vec(boxes: np.ndarray) -> np.ndarray:
w = np.sqrt(boxes[:, 2] * boxes[:, 3])
half_w = w / 2
half_h = boxes[:, 2] / w / 2
xyxy = np.empty_like(boxes)
xyxy[:, 0] = boxes[:, 0] - half_w
xyxy[:, 1] = boxes[:, 1] - half_h
xyxy[:, 2] = boxes[:, 0] + half_w
xyxy[:, 3] = boxes[:, 1] + half_h
return xyxyThere was a problem hiding this comment.
makes sense, though im using it to convert one box at a time and at worst it can run in a loop that has only a few iterations, so performance-wise it should have a big impact, but then you can't input several boxes
| from numpy.typing import NDArray | ||
|
|
||
|
|
||
| class KalmanFilter: |
There was a problem hiding this comment.
Just so I understand. Why do we have specific SORTKalmanBoxTracker and ByteTrackKalmanBoxTracker per tracker? And here KalmanFilter is placed in general directory. The docstring looks like it would match per tracker implementation.
There was a problem hiding this comment.
we discussed it briefly over slack, this is one of the changes i wanted to add for easier scalability of the code. The previous trackers built all in the tracklet class (which are called like ByteTrackKalmanBoxTracker). What i did here is decoupling the KF functionalities from the tracklet. Now the tracklet consumes a KF and makes it nicer to code the different tracklets. We should maybe rewrite the other trackers in this style
trackers/core/ocsort/utils.py
Outdated
| # Copyright (c) 2026 Roboflow. All Rights Reserved. | ||
| # Licensed under the Apache License, Version 2.0 [see LICENSE for details] | ||
| # ------------------------------------------------------------------------ | ||
| # Copied from OC-SORT https://github.com/noahcao/OC_SORT/ |
There was a problem hiding this comment.
I would make it Modified from C-SORT https://github.com/noahcao/OC_SORT/ and invest in code quality of those utils.
…w/trackers into feat/core/ocsort-release
…d update also when unmatched detections = 0
|
@AlexBodner I just merged big PR adding support for |
docs/trackers/ocsort.md
Outdated
|
|
||
| ## API | ||
|
|
||
| ::: trackers.core.ocsort.tracker.OCSORTTracker |
There was a problem hiding this comment.
This should be moved to API Reference docs tab. http://trackers.roboflow.com/develop/api/trackers/
trackers/core/ocsort/utils.py
Outdated
| from trackers.core.ocsort.tracklet import OCSORTTracklet | ||
|
|
||
|
|
||
| def speed_direction(bbox1: np.ndarray, bbox2: np.ndarray) -> np.ndarray: |
There was a problem hiding this comment.
All the functions that we want to keep privat should be prefixed with _. Most of those utils should be private.
…w/trackers into feat/core/ocsort-release
3 participants
What does this PR do?
Adding OC-SORT tracker
Type of Change