add pose tracking demo #380
Conversation
|
|
Codecov Report
@@ Coverage Diff @@
## master #380 +/- ##
==========================================
+ Coverage 81.81% 82.84% +1.02%
==========================================
Files 118 121 +3
Lines 7766 7945 +179
Branches 1240 1265 +25
==========================================
+ Hits 6354 6582 +228
+ Misses 1158 1102 -56
- Partials 254 261 +7
Flags with carried forward coverage won't be shown. Click here to find out more.
Continue to review full report at Codecov.
|
|
Need unittests and README like https://github.com/open-mmlab/mmpose/blob/master/demo/2d_hand_demo.md |
mmpose/apis/inference_tracking.py
Outdated
| boxA_area = (boxA[2] - boxA[0] + 1) * (boxA[3] - boxA[1] + 1) | ||
| boxB_area = (boxB[2] - boxB[0] + 1) * (boxB[3] - boxB[1] + 1) | ||
|
|
||
| iou = inter_area / float(boxA_area + boxB_area - inter_area) |
There was a problem hiding this comment.
Any chance that the denominator equals 0?
There was a problem hiding this comment.
Will this function be used elsewhere?
There was a problem hiding this comment.
Do not worry as boxA_area >= inter_area & boxB_area >= inter_area. The denominator equals 0 only is boxA_area = boxA_area = 0, which is impossible for mmdet to predict such a person bbox.
There was a problem hiding this comment.
Do we need assertion that boxA[2] > boxA[0] and boxA[3] > boxA[1] ? @innerlee
There was a problem hiding this comment.
If a zero-area bbox be considered an error, we can add assertions.
If it is allowed, then we may add a max(1, blablha...) at the denominator.
BTW what's the difference between box and bbox?
There was a problem hiding this comment.
I think adding assertions should be better. And one suggestion is that the previous steps may also consider this kind of error (box[2] <= box[0]). In detail, if such a bbox is given, function inference_top_down_pose_model may have some problems. We may also consider to add assertions in function _xyxy2xywh or anywhere suitable.
As for the difference box and bbox, function _compute_iou is more general and can compute iou for any two boxes. I use box because I do not want to emphasize they are person bounding boxes. However, we may not consider other kinds of boxes and I will use bbox to avoid confusion.
If you guys agree with my opinions, I will make these two changes very soon. @innerlee @jin-s13
There was a problem hiding this comment.
More checks are good. One thing to take care is that we
- use assertion only when the it is an error, because a violation would stop the running program, and any unsaved computations would got lost.
- If the case is not critical, then a warning is good enough.
- If it is neglectable, we can simply ignore it and replace with safe neutral values
I vote for bbox.
|
Dear luminxu, |
Get track id using greedy bbox IOU tracking & Visualize persons according to track id.