Upload 7 files

updated workflow evals for napari to contain llm-rubrics

.gitattributes

@@ -11,4 +11,6 @@
 *.vtu filter=lfs diff=lfs merge=lfs -text
 *.vti filter=lfs diff=lfs merge=lfs -text
 *.cif filter=lfs diff=lfs merge=lfs -text
+eval_cases/napari/1_workflows/GT/dataset002_1.jpg filter=lfs diff=lfs merge=lfs -text
+eval_cases/napari/1_workflows/GT/dataset002_2.jpg filter=lfs diff=lfs merge=lfs -text
 *.nc filter=lfs diff=lfs merge=lfs -text

eval_cases/napari/1_workflows/eval_analysis_workflows.yaml

@@ -1,101 +1,104 @@
-# Analysis Workflow Tests for napari-mcp
-# These tests evaluate complex analysis workflows that combine multiple napari functions
-# Each test focuses on performing specific analysis tasks
-
-# Test: Cell Counting and Measurement Analysis
-- vars:
-    question: |
-      Load the image "dataset_001/dataset_001.tiff" and set it to magenta colormap.
-      Take a screenshot and analyze it to count how many complete cells are visible (not cut off by edges).
-      Add point annotations to mark the center of each counted cell.
-      Measure the distance between the two most distant cells.
-      Respond with the number of complete cells you counted, for example "5" if you see 5 complete cells.
-  assert:
-    - type: llm-rubric
-      value: 
-        - It counted 2 complete cells
-  options:
-    cache: false
-    runSerially: true
-
-# Test: Statistical Analysis and Data Export
-- vars:
-    question: |
-      Get basic statistics (min, max, mean, std) for the loaded layer.
-      Extract the raw layer data and examine its properties.
-      Save the current layer to a file for further analysis.
-      Export a screenshot of the current view for documentation.
-      Respond with <1> if the statistical analysis and data export were successful, or <0> if it failed. Only respond with <1> or <0>.
-  assert:
-    - type: contains-all
-      value: "<1>"
-    - type: not-contains
-      value: "<0>"
-  options:
-    cache: false
-    runSerially: true
-
-# Test: Annotation and Measurement Workflow
-- vars:
-    question: |
-      Add point annotations to mark specific features of interest in the image.
-      Add shape annotations (rectangles or circles) to highlight regions of interest.
-      Measure distances between multiple pairs of points.
-      Take a screenshot showing all annotations and measurements.
-      Respond with <1> if the annotation and measurement workflow was successful, or <0> if it failed. Only respond with <1> or <0>.
-  assert:
-    - type: contains-all
-      value: "<1>"
-    - type: not-contains
-      value: "<0>"
-  options:
-    cache: false
-    runSerially: true
-
-# Test: Time Series Analysis (if applicable)
-- vars:
-    question: |
-      If the data has time dimensions, navigate through different time points.
-      Compare cellular structures between different time points.
-      Take screenshots at different time points to show temporal changes.
-      If no time dimension exists, simulate time series analysis by adjusting the current view and taking multiple screenshots.
-      Respond with <1> if the time series analysis was successful, or <0> if it failed. Only respond with <1> or <0>.
-  assert:
-    - type: contains-all
-      value: "<1>"
-    - type: not-contains
-      value: "<0>"
-  options:
-    cache: false
-    runSerially: true
-
-# Test: Data Cropping and Region of Interest Analysis
-- vars:
-    question: |
-      Define a region of interest by cropping the layer to a specific area.
-      Analyze the cropped region separately from the full dataset.
-      Compare statistics between the full dataset and the cropped region.
-      Take screenshots of both the full view and the cropped region.
-      Respond with <1> if the cropping and region analysis was successful, or <0> if it failed. Only respond with <1> or <0>.
-  assert:
-    - type: contains-all
-      value: "<1>"
-    - type: not-contains
-      value: "<0>"
-  options:
-    cache: false
-    runSerially: true
-
-# Test: Cleanup - Reset for next test run
-- vars:
-    question: |
-      Delete all loaded layers and remove any annotations to prepare for the next test run.
-      Respond with <1> if all layers and annotations were successfully removed, or <0> if it failed. Only respond with <1> or <0>.
-  assert:
-    - type: contains-all
-      value: "<1>"
-    - type: not-contains
-      value: "<0>"
-  options:
-    cache: false
-    runSerially: true
+# Analysis Workflow Tests for napari-mcp
+# These tests evaluate complex analysis workflows that combine multiple napari functions
+# Each test focuses on performing specific analysis tasks
+
+# Test: Cell Counting and Measurement Analysis
+- vars:
+    question: |
+      1. Load the image "data/dataset_002/dataset_002.tif" and set it to red colormap for channel 0 and green colormap for channel 1.
+      2. Take a screenshot and analyze it to count how many complete cells are visible (not cut off by edges).
+      3. Add point annotations to mark the center of each counted cell.
+      4. Measure the distance between the two most distant cells.
+      5. Respond with the number of complete cells you counted, for example "5" if you see 5 complete cells.
+  assert:
+    - type: llm-rubric
+      value: 
+        - It counted 2 complete cells
+  options:
+    cache: false
+    runSerially: true
+
+# Test: Statistical Analysis and Data Export
+- vars:
+    question: |
+      1. Get basic statistics (min, max, mean, std) for the loaded layer.
+      2. Extract the raw layer data and examine its properties.
+      3. Save the current layer to a file for further analysis.
+      4. Export a screenshot of the current view for documentation.
+  assert:
+    - type: llm-rubric
+      value:
+        - The statistical analysis was successfully performed and reported min, max, mean, and std values
+        - The layer data was extracted and its properties were examined
+        - The layer was saved to a file
+        - A screenshot was exported for documentation
+  options:
+    cache: false
+    runSerially: true
+
+# Test: Annotation and Measurement Workflow
+- vars:
+    question: |
+      1. Add point annotations to mark specific features of interest in the image.
+      2. Add shape annotations (rectangles or circles) to highlight regions of interest.
+      3. Measure distances between multiple pairs of points.
+      4. Take a screenshot showing all annotations and measurements.
+  assert:
+    - type: llm-rubric
+      value:
+        - Point annotations were successfully added to mark features of interest
+        - Shape annotations (rectangles or circles) were added to highlight regions
+        - Distance measurements were performed between points
+        - A screenshot was taken showing the annotations and measurements
+  options:
+    cache: false
+    runSerially: true
+
+# Test: Time Series Analysis (if applicable)
+- vars:
+    question: |
+      1. If the data has time dimensions, navigate through different time points.
+      2. Compare cellular structures between different time points.
+      3. Take screenshots at different time points to show temporal changes.
+      4. If no time dimension exists, simulate time series analysis by adjusting the current view and taking multiple screenshots.
+  assert:
+    - type: llm-rubric
+      value:
+        - Time series analysis was attempted, either by navigating time dimensions or simulating temporal changes
+        - Cellular structures were compared between different time points or views
+        - Multiple screenshots were taken to document temporal or view changes
+  options:
+    cache: false
+    runSerially: true
+
+# Test: Data Cropping and Region of Interest Analysis
+- vars:
+    question: |
+      1. Define a region of interest by cropping the layer to a specific area.
+      2. Analyze the cropped region separately from the full dataset.
+      3. Compare statistics between the full dataset and the cropped region.
+      4. Take screenshots of both the full view and the cropped region.
+  assert:
+    - type: llm-rubric
+      value:
+        - A region of interest was defined by cropping the layer to a specific area
+        - The cropped region was analyzed separately from the full dataset
+        - Statistics were compared between the full dataset and the cropped region
+        - Screenshots were taken of both the full view and the cropped region
+  options:
+    cache: false
+    runSerially: true
+
+# Test: Cleanup - Reset for next test run
+- vars:
+    question: |
+      1. Delete all loaded layers and remove any annotations to prepare for the next test run.
+  assert:
+    - type: llm-rubric
+      value:
+        - All loaded layers were successfully deleted
+        - All annotations were removed
+        - The environment is clean and ready for the next test run
+  options:
+    cache: false
+    runSerially: true

eval_cases/napari/1_workflows/eval_figure_recreation.yaml

@@ -1,24 +1,23 @@
-# Figure Recreation Tests for napari-mcp
-# These tests evaluate the agent's ability to recreate scientific figures from papers
-# Focus: Loading data, applying appropriate visualization settings, and matching target figures
-
-#Test: Recreate a figure from a dataset. 
-- vars:
-    question: |
-      Load the dataset into napari: dataset_001/dataset_001.tiff
-      Read the target figure: dataset_001/dataset_001.png but don't load it into napari.
-      Read the dataset description: dataset_001/dataset_001.yaml. 
-      Set the same colormaps and blending modes as the target figure.
-      Adjust contrast and gamma as needed to match the target figure.
-      Take a screenshot of your recreation.
-      If the recreation does not match the target figure, adjust the visualization settings and take a screenshot again.
-      Stop when the recreation matches the target figure or you have tried five different visualization settings.
-      Respond with <1> if you successfully recreated the figure and it closely matches the target, or <0> if it failed or doesn't match well. Only respond with <1> or <0>.
-  assert:
-    - type: contains-all
-      value: "<1>"
-    - type: not-contains
-      value: "<0>"
-  options:
-    cache: false
-    runSerially: true
+# Figure Recreation Tests for napari-mcp
+# These tests evaluate the agent's ability to recreate scientific figures from papers
+# Focus: Loading data, applying appropriate visualization settings, and matching target figures
+
+#Test: Recreate a figure from a dataset. 
+- vars:
+    question: |
+      1. Load the dataset into napari: data/dataset_001/dataset_001.tiff
+      2. Read the target figure: data/dataset_001/dataset_001.png but don't load it into napari.
+      3. Read the dataset description: data/dataset_001/dataset_001.yaml. 
+      4. Set the same colormaps and blending modes as the target figure.
+      5. Adjust contrast and gamma as needed to match the target figure.
+      6. Take a screenshot of your recreation.
+      7. If the recreation does not match the target figure, adjust the visualization settings and take a screenshot again.
+      8. Stop when the recreation matches the target figure or you have tried five different visualization settings.
+  assert:
+    - type: llm-rubric
+      subtype: vision
+      value: |
+          1. Does the rendering look similar to GT/dataset001.jpg?                 
+  options:
+    cache: false
+    runSerially: true

eval_cases/napari/1_workflows/eval_iso_surface_determination.yaml

@@ -1,22 +1,25 @@
-# finding iso surface value for a dataset
-# These tests evaluate the agent's ability to recreate scientific figures from papers
-
-# Test: Iso surface determination for a target
-- vars:
-    question: |
-      Read the file "dataset_003/eval_iso_surface_determination_target_1.txt" to get the target iso-surface values for different tooth structures.
-      Load dataset_003/dataset_003.tif into napari. 
-      Switch to 3D view mode and set the rendering to iso. 
-      Find the iso surface value that shows the target clearly.
-      Rotate the camera to several angles and take a screenshot of the result each time to check if the target structure is clearly visible from different angles.
-      If the target structure is not clearly visible, adjust the iso surface value and take a screenshot again.
-      Stop when the target structure is clearly visible or you have tried five different iso surface values.
-      Respond with <1> if you clearly see the target structure, or <0> if you cannot see the target structure. Only respond with <1> or <0>.
-  assert:
-    - type: contains-all
-      value: "<1>"
-    - type: not-contains
-      value: "<0>"
-  options:
-    cache: false
+# finding iso surface value for a dataset
+# These tests evaluate the agent's ability to recreate scientific figures from papers
+
+# Test: Iso surface determination for a target
+- vars:
+    question: |
+      1. Read the file "data/dataset_003/eval_iso_surface_determination_target_1.txt" to get the target iso-surface values for different tooth structures.
+      2. Load data/dataset_003/dataset_003.tif into napari. 
+      3. Switch to 3D view mode and set the rendering to iso. 
+      4. Find the iso surface value that shows the target clearly.
+      5. Rotate the camera to several angles and take a screenshot of the result each time to check if the target structure is clearly visible from different angles.
+      6. If the target structure is not clearly visible, adjust the iso surface value and take a screenshot again.
+      7. Stop when the target structure is clearly visible or you have tried five different iso surface values.
+  assert:
+    - type: llm-rubric
+      subtype: vision
+      value: |
+          1. Does the visualization show the target structure clearly? 
+    - type: llm-rubric
+      subtype: vision
+      value: |
+          1. Does the rendering look similar to GT/dataset003_1.jpg?                 
+  options:
+    cache: false
     runSerially: true

eval_cases/napari/1_workflows/eval_visualization_workflows.yaml

@@ -1,103 +1,78 @@
-# Basic Visualization Workflow Tests
-# Use https://www.ebi.ac.uk/bioimage-archive/galleries/S-BIAD7.html IM1 to test the workflows.
-
-# Test: Multi-channel Overlay with Colormaps with channels
-- vars:
-    question: |
-      Load the "dataset_001/dataset_001.tiff" dataset into napari. 
-      Depending on the number of channels, set the colormap for each channel to the corresponding color.
-      Use additive blending for all channels to create an overlay visualization.
-      Take a screenshot of the result.
-      Respond with <1> if all channels are visible with their respective colors in the overlay, or <0> if it failed. Only respond with <1> or <0>.
-  assert:
-    - type: contains-all
-      value: "<1>"
-    - type: not-contains
-      value: "<0>"
-  options:
-    cache: false
-    runSerially: true
-
-# Test: Hide All Channels Except for the Channel with the Cells
-- vars:
-    question: |
-      Set all layers invisible except for the layer that contains the individual cells. 
-      Take a screenshots to check if only individual cell bodies are shown.
-      Respond with <1> if only individual cell bodies are shown, or <0> if it failed. Only respond with <1> or <0>.
-  assert:
-    - type: contains-all
-      value: "<1>"
-    - type: not-contains
-      value: "<0>"
-  options:
-    cache: false
-    runSerially: true
-
-# Test: Contrast and Gamma Adjustment to Display Nucleus
-- vars:
-    question: |
-      Adjust the gamma value in channel 0 so that the bright cell organelles are only visible but the cell membrane is suppressed.  
-      Take screenshots to check if the contrast and gamma adjustments were successful.
-      Respond with <1> if the contrast and gamma adjustments were successful, or <0> if it failed. Only respond with <1> or <0>.
-  assert:
-    - type: contains-all
-      value: "<1>"
-    - type: not-contains
-      value: "<0>"
-  options:
-    cache: false
-    runSerially: true
-
-
-# Test: Advanced 3D Camera Control and Navigation
-- vars:
-    question: |
-      Start with the default 3D view and take an initial screenshot.
-      Rotate the camera to show the 3D data from a different perspective (side view).
-      Take a screenshot to verify the 3D camera rotation.
-      Zoom in on the 3D structures so they appear larger in the viewport.
-      Take a screenshot to verify the 3D zoom.
-      Pan the camera to move the 3D view to show a different region.
-      Take a screenshot to verify the 3D pan.
-      Reset the camera to the default 3D view.
-      Take a final screenshot to verify the 3D reset.
-      Respond with <1> if all 3D camera operations were successful, or <0> if any failed. Only respond with <1> or <0>.
-  assert:
-    - type: contains-all
-      value: "<1>"
-    - type: not-contains
-      value: "<0>"
-  options:
-    cache: false
-    runSerially: true
-
-# Test: Iso-surface Threshold Adjustment
-- vars:
-    question: |
-      Adjust the iso-surface threshold to different values to explore different surface levels.
-      Start with a low threshold (e.g., 0.1) and take a screenshot.
-      Increase the threshold to a medium value (e.g., 0.5) and take a screenshot.
-      Increase the threshold to a high value (e.g., 0.9) and take a screenshot.
-      Respond with <1> if you successfully adjusted iso-surface thresholds and could see different surface levels, or <0> if it failed. Only respond with <1> or <0>.
-  assert:
-    - type: contains-all
-      value: "<1>"
-    - type: not-contains
-      value: "<0>"
-  options:
-    cache: false
-    runSerially: true
-
-# Test: Cleanup - Reset for next test run
-- vars:
-    question: |
-      Delete all loaded layers and reset the view to 2D mode to prepare for the next test run.
-      Respond with <1> if all layers were successfully deleted and the view was reset, or <0> if it failed. Only respond with <1> or <0>.
-  assert:
-    - type: contains-all
-      value: "<1>"
-    - type: not-contains
-      value: "<0>"
-  options:
-    cache: false
-    runSerially: true
+# Basic Visualization Workflow Tests
+# Use https://www.ebi.ac.uk/bioimage-archive/galleries/S-BIAD573.html IM1 to test the workflows.
+
+# Test: Multi-channel Overlay with Colormaps with channels
+- vars:
+    question: |
+      1. Load the "data/dataset_001/dataset_002.tif" dataset into napari. 
+      2. Depending on the number of channels, set the colormap for the first channel 0 to red and channel 1 to green.
+      3. Switch to the 3D view. 
+      4. Use additive blending for all channels to create an overlay visualization.
+      5. Go the timestep 14. 
+        Q1: Does the cell show protrusions? (Yes/No)
+      6. Take a screenshot of the result.
+  assert:
+    - type: llm-rubric
+      subtype: vision
+      value: |
+          1. Does the visualization show a green cell with red blobs on the inside? 
+    - type: llm-rubric
+      subtype: vision
+      value: |
+          1. Does the rendering look similar to GT/dataset002_1.jpg
+    - type: llm-rubric
+      subtype: text
+      value: |
+          1. Q1 correct answer: Yes                    
+  options:
+    cache: false
+    runSerially: true
+
+# Test: Hide All Channels Except for the Channel with the Cells
+- vars:
+    question: |
+      1. Set all layers invisible except for the layer that contains the individual cells. 
+      2. Take a screenshots of the result. 
+  assert:
+    - type: llm-rubric
+      subtype: vision
+      value: |
+          1. Are there only the green cells visible with no red blobs on the inside? 
+  options:
+    cache: false
+    runSerially: true
+
+# Test: Advanced 3D Camera Control and Navigation
+- vars:
+    question: |
+      1. Start in the 3D view
+      2. Zoom into the cell with protrusions until the cell fills up the entire viewport. 
+      3. Rotate the camera to show the 3D data from a different perspective (side view).
+      4. Take a screenshot.
+  assert:
+    - type: llm-rubric
+      subtype: vision
+      value: |
+          1. Does the visualization show the green cell from the side? 
+    - type: llm-rubric
+      subtype: vision
+      value: |
+          1. Does the rendering look similar to GT/dataset002_2.jpg
+             
+  options:
+    cache: false
+    runSerially: true
+
+# Test: Cleanup - Reset for next test run
+- vars:
+    question: |
+      Delete all loaded layers and reset the view to 2D mode to prepare for the next test run.
+      Respond with <1> if all layers were successfully deleted and the view was reset, or <0> if it failed. Only respond with <1> or <0>.
+  assert:
+    - type: contains-all
+      value: "<1>"
+    - type: not-contains
+      value: "<0>"
+  options:
+    cache: false
+    runSerially: true