Mercurial > repos > goeckslab > image_learner

comparison html_structure.py @ 19:c460abae83eb draft default tip

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
planemo upload for repository https://github.com/goeckslab/gleam.git commit b47f0fd63d8d5d18d602d45bb21ebbe36ba4fcfe
author goeckslab
date Thu, 18 Dec 2025 16:59:58 +0000
parents db9be962dc13
children
comparison
equal deleted inserted replaced
18:bbf30253c99f 19:c460abae83eb
229 return """ 229 return """
230 <!DOCTYPE html> 230 <!DOCTYPE html>
231 <html> 231 <html>
232 <head> 232 <head>
233 <meta charset="UTF-8"> 233 <meta charset="UTF-8">
234 <title>Galaxy-Ludwig Report</title> 234 <title>Image Learner Report</title>
235 <style> 235 <style>
236 body { 236 body {
237 font-family: Arial, sans-serif; 237 font-family: Arial, sans-serif;
238 margin: 0; 238 margin: 0;
239 padding: 20px; 239 padding: 20px;
717 'to account for class imbalance.</li>' 717 'to account for class imbalance.</li>'
718 ' <li><strong>Recommendation/Ranking:</strong> Use <strong>Hits at K</strong> for retrieval tasks.</li>' 718 ' <li><strong>Recommendation/Ranking:</strong> Use <strong>Hits at K</strong> for retrieval tasks.</li>'
719 ' <li><strong>Detailed Analysis:</strong> Use <strong>Confusion Matrix stats</strong> ' 719 ' <li><strong>Detailed Analysis:</strong> Use <strong>Confusion Matrix stats</strong> '
720 'for class-wise performance in classification.</li>' 720 'for class-wise performance in classification.</li>'
721 ' </ul>' 721 ' </ul>'
722 ' <h3>11) Grad-CAM Heatmaps (When Available)</h3>'
723 ' <p><strong>Paper:</strong> <em>Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localization</em> '
724 '(Selvaraju, Cogswell, Das, Vedantam, Parikh, Batra; ICCV 2017).</p>'
725 ' <p><strong>What it shows:</strong> A heatmap highlighting image regions that most influenced the model’s prediction '
726 'for a small subset of evaluation samples (we prefer the test split when available).</p>'
727 ' <p><strong>How it is computed (high level):</strong> We use the encoder’s preprocessing (resize + normalization), '
728 'take activations from the last convolution layer, weight them by globally-averaged gradients of the target logits, apply ReLU, '
729 'upsample to input resolution, and overlay on the input image.</p>'
730 ' <p><strong>Availability:</strong> Only supported for convolutional encoders. Models without convolution layers may not '
731 'produce Grad-CAM outputs.</p>'
722 ' </div>' 732 ' </div>'
723 ' </div>' 733 ' </div>'
724 '</div>' 734 '</div>'
725 ) 735 )
726 736