New alerte strategy #164
Merged
New alerte strategy #164
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Nice of you to open a PR 🙏 |
Merged
MateoLostanlen
added a commit
that referenced
this pull request
Aug 16, 2023
* update header (#149) * fix black (#150) * Update README.md (#151) * Promtail (#152) * Update of the main branch before the new deployment in Ardeche (#145) * ci on develop (#134) * update develop tour (#135) * fix args (#136) * fix: Fixed script arg name in src/run.py (#138) * clean backup by size (#141) * clean backup by size * black * missing deps * drop function * drop function 2 * Day only (#142) * sunset_sunrise script * update output file path * check if day time * put back real api * style * use datetime timedelta * Make all params availables from run (#144) * make all params available in run * make params availables * put back api * style * put back alert_relaxation to 2 * Yolov5 (#143) * switch to yolov5 * missing import * style * fix tests * unused import * update readme * letterbox transform * do not resize with pillow before pred * style * downgrad opencv * wrong img name * long line * missing deps * model path * header * lib for opencv * create model folder * update init * remove hub deps * update threshold * feat: Added promtail service * feat: Added promtail config file * feat: Renamed variable * feat: Renamed vars * docs: Updated setup explanation * feat: Added restart option --------- Co-authored-by: Mateo <[email protected]> * analyze stream timeout (#153) * update docker img (#154) * add utc (#155) * fix style (#158) * Ir day night (#159) * set margin to 0 * drop utc * add ir strategy * update docstring * fix style * fix missing undefinded params (#161) * Add bbox (#162) * add bbox * fix inference * style * fix vision * style vision * fix engine * keep all preds * speed up * pass dummy loc * switch to v8 * fix tests * test * drop dummy test * new api version * unsed import * install git * code quality * use bbox branch * use apt get * alert relaxation 3 * New alerte strategy (#164) * new alert system * fix unitests * Occlusion mask (#165) * add occlusion mask * trest mask * missing args * use cam_key * update test * add a test * clip values * install from main (#167) * limit bbox size (#166) * limit bbox size * style * use 4 frames (#168) * don't send bbox (#169) * prevent empty (#171) --------- Co-authored-by: fe51 <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Hello,
The counter we're currently using poses a problem if we have the following prediction sequence (0: no_fire, 1:fire):
1 0 1 1 0 1 1 1 1
We'll start transmitting images from frame 7 onwards, although the previous images are probably interesting.
What's more, we're not using our model's location information, so two FPs in different places will increase the counter together.
I therefore propose a new strategy, combining the results of N consecutive frames. As shown in the schema below, the idea is to perform an NMS on all the boxes amoung the N frames, then for each resulting box, sum up the score of each box with an intersection >0. In the schema, we sum up all the red boxes to obtain a score of 1.21.
The new detection threshold is conf_th * N, i.e. here 0.25 x 4 = 1. Only the red box is retained.
We then send the 4 frames used to make the prediction, in order not to lose any image.
NB: in this case, at frame 4, the old counter would be at 1, meaning that many images are lost.