Detection #16
Detection #16
Conversation
NetZissou
commented
Jul 21, 2025
NetZissou
commented
- Base YOLO Detector
- Camera-trap animal detection (closes Animal Detection #7 )
- Face detection (closes Face Detection #6 )
…the original image
Face Detection
Merge Animal Detection
NetZissou
added
documentation
…ated authors list; updated project URL sections
| Homepage = "https://github.com/Imageomics/hpc-inference" | ||
| Documentation = "https://imageomics.github.io/hpc-inference/" | ||
| Repository = "https://github.com/Imageomics/hpc-inference" | ||
| Changelog = "https://github.com/Imageomics/hpc-inference/blob/main/CHANGELOG.md" |
There was a problem hiding this comment.
| Changelog = "https://github.com/Imageomics/hpc-inference/blob/main/CHANGELOG.md" |
|
|
||
| ### Key Features | ||
|
|
||
| - **MegaDetector Support**: Optimized for Microsoft's MegaDetector models (specifically designed for wildlife camera trap images) |
There was a problem hiding this comment.
Is it that it's specifically designed to be use with Pytorch-Wildlife (https://github.com/microsoft/CameraTraps)?
There was a problem hiding this comment.
I would compare PytorchWildlife to pybioclip. It provides a convenient Python interface for running inference with Megadetector, but it does not address I/O bottlenecks and is not optimized for scaling or for efficiently handling large inference tasks.
The Megadetector model weight files can be downloaded directly from the model zoo. hpc-inference interacts with these .pt model files directly.
There was a problem hiding this comment.
But those come from Megadector?
The issue is specifying specifically what's used, but not misattributing it as "Microsoft's" Megadetector. Earlier code is licensed to Microsoft (see note in repo).
There was a problem hiding this comment.
Got it! We’re using Pytorch-Wildlife-model-weights, which were prepared by the PytorchWildLife team. The correct attribution should therefore go to PytorchWildlife, not Microsoft’s Megadetector.
There was a problem hiding this comment.
I'd just add it to the citation further up (since you cite Megadetector with both already). Also, change:
| - **MegaDetector Support**: Optimized for Microsoft's MegaDetector models (specifically designed for wildlife camera trap images) | |
| - **MegaDetector Support**: Optimized for [Pytorch-Wildlife-model-weights](https://zenodo.org/records/15398270), specifically, the Megadetector models designed for wildlife camera trap images. |
| - JPEG (.jpg, .jpeg) | ||
| - PNG (.png) | ||
| - BMP (.bmp) | ||
| - TIFF (.tif, .tiff) | ||
| - WebP (.webp) |
There was a problem hiding this comment.
| - JPEG (.jpg, .jpeg) | |
| - PNG (.png) | |
| - BMP (.bmp) | |
| - TIFF (.tif, .tiff) | |
| - WebP (.webp) | |
| - JPEG (`.jpg`, `.jpeg`) | |
| - PNG (`.png`) | |
| - BMP (`.bmp`) | |
| - TIFF (`.tif`, `.tiff`) | |
| - WebP (`.webp`) |
|
|
||
| Configuration files provide better reproducibility and easier parameter management. | ||
|
|
||
| You can find documented template for config specification in `configs/animal_detection` |
There was a problem hiding this comment.
| You can find documented template for config specification in `configs/animal_detection` | |
| You can find documented templates for config specification in `configs/animal_detection`. |
| 'class_name': 'animal'}] | ||
| ``` | ||
|
|
||
| Megadetector detected one animal with high confident 0.8981. Let's use the utility function to plot the detection box on top of the original image to validate the detection result: |
There was a problem hiding this comment.
| Megadetector detected one animal with high confident 0.8981. Let's use the utility function to plot the detection box on top of the original image to validate the detection result: | |
| Megadetector detected one animal with high confidence (0.8981). Let's use the utility function to plot the detection box on top of the original image to validate the detection result: |
| title=f"Detections for {image_uuid}" | ||
| ) | ||
| ``` | ||
|  |
There was a problem hiding this comment.
|  | |
|  |
Add in source for original image. May want to put in a caption or something.
