bioimage.io upload: ambitious-kangaroo
Fuente:
Zenodo
Guardado en:
| Autor principal: | |
|---|---|
| Formato: | Recurso digital |
| Publicado: |
Zenodo
2025
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866902335337267200 |
|---|---|
| author | Daniel Franco-Barranco |
| author_facet | Daniel Franco-Barranco |
| contents | <p><a href="" title="https://bioimage.io/#/?id=ambitious-kangaroo">View on bioimage.io</a> # 2D UNETR for cell segmentation in fluorescence </p> <h1>2D UNETR for cell segmentation in fluorescence</h1> <p>Live cell (E. coli) segmentation for fluorescence widefield images..</p> <h2>Training</h2> <p>Complete information on how to train this model can be found in BiaPy documentation, specifically under <a href="https://biapy.readthedocs.io/en/latest/workflows/instance_segmentation.html">instance segmentation workflow</a> description.</p> <h3>Training Data</h3> <ul> <li>Imaging modality: fluorescence widefield</li> <li>Dimensionality: 2D</li> <li>Source: deepbacs ; DOI:10.5281/zenodo.5639253</li> </ul> <h3>Validation</h3> <p>In the case of instance segmentation the following metrics are calculated: - Intersection over Union (IoU), also referred as the <a href="https://en.wikipedia.org/wiki/Jaccard_index">Jaccard index</a>, is essentially a method to quantify the percent of overlap between the target mask and the prediction output. In this workflow, as in BiaPy a botton-down approach is used to generate the final instances, IoU is used to measure the model output with the mask created out of the instances. - Matching metrics, which focus on quantifying correctly predicted instances, transforming instance segmentation results into an object detection framework. In this paradigm, the emphasis shifts from uniquely labeled instances to detecting the presence or absence of instances. This transformation is achieved by establishing a criterion for instance overlap, commonly measured through IoU. Unlike traditional segmentation evaluations that rely on nuanced pixel-level overlaps, this approach simplifies assessment by classifying instances as successful true positives (TP) based on a predefined IoU threshold. In BiaPy, the following metrics are implemented: false positives (FP), true positives (TP), false negatives (FN), <a href="https://en.wikipedia.org/wiki/Precision_and_recall">precision</a>, <a href="https://en.wikipedia.org/wiki/Precision_and_recall">recall</a>, <a href="https://en.wikipedia.org/wiki/Accuracy_and_precision#In_binary_classification">accuracy</a>, <a href="https://en.wikipedia.org/wiki/Precision_and_recall#F-measure">F1-score</a>, mean_true_score (mean IoUs of matched true positives but normalized by the total number of GT objects) and panoptic_quality (defined as in Eq. 1 of (Kirillov et al. 'Panoptic Segmentation', CVPR 2019)[https://openaccess.thecvf.com/content_CVPR_2019/papers/Kirillov_Panoptic_Segmentation_CVPR_2019_paper.pdf]).</p> <p>More info in <a href="https://biapy.readthedocs.io/en/latest/workflows/instance_segmentation.html#metrics">BiaPy documentation</a>.</p> <h3>Training Schedule (BiaPy variables)</h3> <h4>Training setup</h4> <p>TRAIN: BATCH_SIZE: 2 ENABLE: True EPOCHS: 1500 LR: 0.0001 LR_SCHEDULER: {'MIN_LR': 1e-05, 'NAME': 'warmupcosine', 'WARMUP_COSINE_DECAY_EPOCHS': 50} OPTIMIZER: ADAMW PATIENCE: 250</p> <h4>Data augmentation</h4> <p>AUGMENTOR: AFFINE_MODE: reflect BRIGHTNESS: True BRIGHTNESS_FACTOR: (-0.2, 0.2) CONTRAST: True CONTRAST_FACTOR: (-0.2, 0.2) ELASTIC: True ENABLE: True HFLIP: True RANDOM_ROT: True VFLIP: True ZFLIP: True ZOOM: True ZOOM_RANGE: (0.8, 1.2)</p> <h2>Contact</h2> <p>For problems with BiaPy library itself checkout our <a href="https://biapy.readthedocs.io/en/latest/get_started/faq.html">FAQ & Troubleshooting section</a>.</p> <p>For questions or issues with this models, please reach out by: - Opening a topic with tags bioimageio and biapy on <a href="https://forum.image.sc/">image.sc</a> - Creating an issue in https://github.com/BiaPyX/BiaPy</p> <p>Model created by: - Daniel Franco-Barranco (github: danifranco)</p> |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_15691456 |
| institution | Zenodo |
| language | |
| publishDate | 2025 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | bioimage.io upload: ambitious-kangaroo Daniel Franco-Barranco backup.bioimage.io bioimage.io bioimage.io:model fluorescence widefield bacillus-subtilis microbiology bacteria 2d pytorch biapy instance-segmentation unetr <p><a href="" title="https://bioimage.io/#/?id=ambitious-kangaroo">View on bioimage.io</a> # 2D UNETR for cell segmentation in fluorescence </p> <h1>2D UNETR for cell segmentation in fluorescence</h1> <p>Live cell (E. coli) segmentation for fluorescence widefield images..</p> <h2>Training</h2> <p>Complete information on how to train this model can be found in BiaPy documentation, specifically under <a href="https://biapy.readthedocs.io/en/latest/workflows/instance_segmentation.html">instance segmentation workflow</a> description.</p> <h3>Training Data</h3> <ul> <li>Imaging modality: fluorescence widefield</li> <li>Dimensionality: 2D</li> <li>Source: deepbacs ; DOI:10.5281/zenodo.5639253</li> </ul> <h3>Validation</h3> <p>In the case of instance segmentation the following metrics are calculated: - Intersection over Union (IoU), also referred as the <a href="https://en.wikipedia.org/wiki/Jaccard_index">Jaccard index</a>, is essentially a method to quantify the percent of overlap between the target mask and the prediction output. In this workflow, as in BiaPy a botton-down approach is used to generate the final instances, IoU is used to measure the model output with the mask created out of the instances. - Matching metrics, which focus on quantifying correctly predicted instances, transforming instance segmentation results into an object detection framework. In this paradigm, the emphasis shifts from uniquely labeled instances to detecting the presence or absence of instances. This transformation is achieved by establishing a criterion for instance overlap, commonly measured through IoU. Unlike traditional segmentation evaluations that rely on nuanced pixel-level overlaps, this approach simplifies assessment by classifying instances as successful true positives (TP) based on a predefined IoU threshold. In BiaPy, the following metrics are implemented: false positives (FP), true positives (TP), false negatives (FN), <a href="https://en.wikipedia.org/wiki/Precision_and_recall">precision</a>, <a href="https://en.wikipedia.org/wiki/Precision_and_recall">recall</a>, <a href="https://en.wikipedia.org/wiki/Accuracy_and_precision#In_binary_classification">accuracy</a>, <a href="https://en.wikipedia.org/wiki/Precision_and_recall#F-measure">F1-score</a>, mean_true_score (mean IoUs of matched true positives but normalized by the total number of GT objects) and panoptic_quality (defined as in Eq. 1 of (Kirillov et al. 'Panoptic Segmentation', CVPR 2019)[https://openaccess.thecvf.com/content_CVPR_2019/papers/Kirillov_Panoptic_Segmentation_CVPR_2019_paper.pdf]).</p> <p>More info in <a href="https://biapy.readthedocs.io/en/latest/workflows/instance_segmentation.html#metrics">BiaPy documentation</a>.</p> <h3>Training Schedule (BiaPy variables)</h3> <h4>Training setup</h4> <p>TRAIN: BATCH_SIZE: 2 ENABLE: True EPOCHS: 1500 LR: 0.0001 LR_SCHEDULER: {'MIN_LR': 1e-05, 'NAME': 'warmupcosine', 'WARMUP_COSINE_DECAY_EPOCHS': 50} OPTIMIZER: ADAMW PATIENCE: 250</p> <h4>Data augmentation</h4> <p>AUGMENTOR: AFFINE_MODE: reflect BRIGHTNESS: True BRIGHTNESS_FACTOR: (-0.2, 0.2) CONTRAST: True CONTRAST_FACTOR: (-0.2, 0.2) ELASTIC: True ENABLE: True HFLIP: True RANDOM_ROT: True VFLIP: True ZFLIP: True ZOOM: True ZOOM_RANGE: (0.8, 1.2)</p> <h2>Contact</h2> <p>For problems with BiaPy library itself checkout our <a href="https://biapy.readthedocs.io/en/latest/get_started/faq.html">FAQ & Troubleshooting section</a>.</p> <p>For questions or issues with this models, please reach out by: - Opening a topic with tags bioimageio and biapy on <a href="https://forum.image.sc/">image.sc</a> - Creating an issue in https://github.com/BiaPyX/BiaPy</p> <p>Model created by: - Daniel Franco-Barranco (github: danifranco)</p> |
| title | bioimage.io upload: ambitious-kangaroo |
| topic | backup.bioimage.io bioimage.io bioimage.io:model fluorescence widefield bacillus-subtilis microbiology bacteria 2d pytorch biapy instance-segmentation unetr |
| url | https://doi.org/10.5281/zenodo.15691456 |