From 7eed607135f87f4cdeeea15d9409bb73474112de Mon Sep 17 00:00:00 2001 From: wilmund Date: Sun, 30 Aug 2026 14:29:15 +0200 Subject: [PATCH] Fix script path and --bRenderDetectionTiles flag in repeat_detection_elimination README --- .../postprocessing/repeat_detection_elimination/README.md | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/megadetector/postprocessing/repeat_detection_elimination/README.md b/megadetector/postprocessing/repeat_detection_elimination/README.md index ca1f7fb9..59ab92b3 100644 --- a/megadetector/postprocessing/repeat_detection_elimination/README.md +++ b/megadetector/postprocessing/repeat_detection_elimination/README.md @@ -45,7 +45,7 @@ So let's assume that: You would run: -`python megadetector/postprocessing/find_repeat_detections.py "c:\my_results.json" --imageBase "c:\my_images" --outputBase "c:\repeat_detection_stuff"` +`python megadetector/postprocessing/repeat_detection_elimination/find_repeat_detections.py "c:\my_results.json" --imageBase "c:\my_images" --outputBase "c:\repeat_detection_stuff"` This script can take a while! Possibly hours if you have millions of images, typically just a few minutes if you only have tens of thousands of images. If you want to test it on just a couple folders first, you can use the `--debugMaxDir` option, to tell the script to only process a certain number of cameras. @@ -132,7 +132,7 @@ After running this process, you still have a .json file in the standard MegaDete When you see a red box on a rock and you don't delete that image, you're making a judgement call that all the other detections that appear in more or less exactly that same spot are also that same rock. But you're always taking a small risk that an animal happened to line itself up just so at some point, i.e. that it lined up exactly with that rock, in which case suppressing the potentially hundreds of repetitions of that rock could also suppress that animal. With a sufficiently high IoU threshold (more on this below, but basically, how similar two boxes need to be to be considered the same) and a sufficiently high occurrence threshold (more in this below, but basically, the number of times a detection has to occur to be "suspicious"), the risk is low. But it's not zero! -Ergo, we've recently added a neat new feature (thanks, [Doantam](https://www.linkedin.com/in/doantam-phan/)!) that lets you visualize a grid of many (possibly all) of the detections that were identical to the one in the red box. This lets you quickly see what you're throwing away when you don't delete one of these images. We will probably make this the default at some point, because it's super-duper-useful, but we don't like to rock the boat. For now, you can enable this with the `--renderDetectionTiles` option. +Ergo, we've recently added a neat new feature (thanks, [Doantam](https://www.linkedin.com/in/doantam-phan/)!) that lets you visualize a grid of many (possibly all) of the detections that were identical to the one in the red box. This lets you quickly see what you're throwing away when you don't delete one of these images. We will probably make this the default at some point, because it's super-duper-useful, but we don't like to rock the boat. For now, you can enable this with the `--bRenderDetectionTiles` option. Here's an example where you can see in just a glance that all 99 instances of this detection are exactly the same bush: @@ -168,7 +168,7 @@ There are a few magic numbers involved in identifying "suspicious" detections. You can run: -`python megadetector/postprocessing/find_repeat_detections.py` +`python megadetector/postprocessing/repeat_detection_elimination/find_repeat_detections.py` ...for a full list of options and documentation for each, but some specific options of interest.