From 137b23fe78a628151639e9ef5f090e24b1ed27c4 Mon Sep 17 00:00:00 2001 From: goddesswarship <103530016+goddesswarship@users.noreply.github.com> Date: Wed, 24 Jun 2026 13:45:14 -0700 Subject: [PATCH 1/3] Update image-analysis-pipeline.md Added explanations for how MiewID and PIE interpret loss functions to find matches. --- docs/introduction/image-analysis-pipeline.md | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/docs/introduction/image-analysis-pipeline.md b/docs/introduction/image-analysis-pipeline.md index 9a8da59c..0affb39f 100644 --- a/docs/introduction/image-analysis-pipeline.md +++ b/docs/introduction/image-analysis-pipeline.md @@ -44,7 +44,7 @@ The following ID algorithms are supported by Wildbook: MIEW-ID (µID) is used to identify individuals. MIEW-ID uses deep learning and can learn what makes images similar or dissimilar (or what differentiates one animal from another). Distinguishing individuals through their unique body markings is a key concept for wildlife conservation. However, from the huge database of wildlife images, only a limited number can be used for individual identification, due to constraints on image quality and viewpoint. MIEW-ID can identify individuals from their unique body markings across real world photographic conditions. -MIEW-ID learns embeddings for images from the database. Embeddings are the unique markings that represent individuals. When new images are analyzed, their embeddings are matched against those in the database. As an added benefit, MIEW-ID is able generate visualizations of matched features, providing import inspectability inside its neural network. +MIEW-ID learns embeddings for images from the database. Embeddings are a fixed-length list of numbers that represent the unique markings in individuals. When new images are analyzed, their embeddings are interpreted as angles on a sphere and matched against the database. As an added benefit, MIEW-ID is able generate visualizations of matched features, providing import inspectability inside its neural network. ![](../assets/images/beluga-gradCAM.png) @@ -56,7 +56,7 @@ MIEW-ID can be trained on a per-species or multi-species basis and has been succ **Pose Invariant Embeddings** (PIE) is used to identify individuals. PIE uses a type of machine learning known as deep learning. This means it can learn what makes images similar or dissimilar (or what differentiates one animal from another). Distinguishing individuals through their unique body markings is a key concept for wildlife conservation. However, from the huge database of wildlife images, only a limited number can be used for individual identification, due to constraints on image quality and viewpoint. PIE can identify individuals from their unique body markings, regardless of quality or angles. -PIE learns embeddings for images from the database. Embeddings are the unique markings that represent individuals. When new images are analyzed, their embeddings are matched against those in the database. +PIE learns embeddings for images from the database. Embeddings are a fixed-length list of numbers that represent the unique markings in individuals. New images are analyzed as triplets: your anchor or source image, a positive representing the same individual, and a negative representing a different individual. Potential matches place the anchor closer to its positive than to its negative. PIE can be trained on a per-species basis. Currently, Wild Me has generated separate PIE models for different species including hyenas, leopards, manta rays, humpback whales, right whales, bottlenose dolphins, and orcas. @@ -114,4 +114,4 @@ Even machine learning makes mistakes. Users can use [Manual Annotation](../data/ ### I am a software developer of ML engineer. How can I learn more about WBIA? -Here is a link to [Wildbook Image Analysis Overview.](index.md) \ No newline at end of file +Here is a link to [Wildbook Image Analysis Overview.](index.md) From d6af29718f2cfa29ec806e01be5134a8747bfaca Mon Sep 17 00:00:00 2001 From: goddesswarship <103530016+goddesswarship@users.noreply.github.com> Date: Wed, 24 Jun 2026 13:46:57 -0700 Subject: [PATCH 2/3] Update image-analysis-pipeline.md --- docs/introduction/image-analysis-pipeline.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/docs/introduction/image-analysis-pipeline.md b/docs/introduction/image-analysis-pipeline.md index 0affb39f..49b9c788 100644 --- a/docs/introduction/image-analysis-pipeline.md +++ b/docs/introduction/image-analysis-pipeline.md @@ -44,7 +44,7 @@ The following ID algorithms are supported by Wildbook: MIEW-ID (µID) is used to identify individuals. MIEW-ID uses deep learning and can learn what makes images similar or dissimilar (or what differentiates one animal from another). Distinguishing individuals through their unique body markings is a key concept for wildlife conservation. However, from the huge database of wildlife images, only a limited number can be used for individual identification, due to constraints on image quality and viewpoint. MIEW-ID can identify individuals from their unique body markings across real world photographic conditions. -MIEW-ID learns embeddings for images from the database. Embeddings are a fixed-length list of numbers that represent the unique markings in individuals. When new images are analyzed, their embeddings are interpreted as angles on a sphere and matched against the database. As an added benefit, MIEW-ID is able generate visualizations of matched features, providing import inspectability inside its neural network. +MIEW-ID learns embeddings for images from the database. Embeddings are a fixed-length list of numbers that represent the unique markings in individuals. When new images are analyzed, their embeddings are interpreted as angles on a sphere and matched against the database. As an added benefit, MIEW-ID is able to generate visualizations of matched features, providing import inspectability inside its neural network. ![](../assets/images/beluga-gradCAM.png) From e94f411755b120c381785f52c4e862972d44192d Mon Sep 17 00:00:00 2001 From: goddesswarship <103530016+goddesswarship@users.noreply.github.com> Date: Thu, 27 Aug 2026 11:36:44 -0700 Subject: [PATCH 3/3] Update image-analysis-pipeline.md Fixed two typos --- docs/introduction/image-analysis-pipeline.md | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/docs/introduction/image-analysis-pipeline.md b/docs/introduction/image-analysis-pipeline.md index 49b9c788..42266bde 100644 --- a/docs/introduction/image-analysis-pipeline.md +++ b/docs/introduction/image-analysis-pipeline.md @@ -44,7 +44,7 @@ The following ID algorithms are supported by Wildbook: MIEW-ID (µID) is used to identify individuals. MIEW-ID uses deep learning and can learn what makes images similar or dissimilar (or what differentiates one animal from another). Distinguishing individuals through their unique body markings is a key concept for wildlife conservation. However, from the huge database of wildlife images, only a limited number can be used for individual identification, due to constraints on image quality and viewpoint. MIEW-ID can identify individuals from their unique body markings across real world photographic conditions. -MIEW-ID learns embeddings for images from the database. Embeddings are a fixed-length list of numbers that represent the unique markings in individuals. When new images are analyzed, their embeddings are interpreted as angles on a sphere and matched against the database. As an added benefit, MIEW-ID is able to generate visualizations of matched features, providing import inspectability inside its neural network. +MIEW-ID learns embeddings for images from the database. Embeddings are a fixed-length list of numbers that represent the unique markings in individuals. When new images are analyzed, their embeddings are interpreted as angles on a sphere and matched against the database. As an added benefit, MIEW-ID is able to generate visualizations of matched features, providing important inspectability inside its neural network. ![](../assets/images/beluga-gradCAM.png) @@ -112,6 +112,6 @@ Here is a link to an [example training video](https://www.youtube.com/watch?v=qD Even machine learning makes mistakes. Users can use [Manual Annotation](../data/manual-annotation-beta.md) if detection doesn’t find an animal. -### I am a software developer of ML engineer. How can I learn more about WBIA? +### I am a software developer or ML engineer. How can I learn more about WBIA? Here is a link to [Wildbook Image Analysis Overview.](index.md)