Image visualizations

Requirements

Requires the image-dataset-converter-imgvis module.

Plugins

Annotation overlays (image classification)

Download the 17 flowers image classification dataset and extract it.

Having annotations separate from the images is a necessity when training models, but it can be a hindrance when trying to inspect the data. Adding overlays with the annotations is therefore a useful step sometimes.

The following command using the add-annotation-overlay-ic filter adds the labels to the images and just outputs these modified images:

idc-convert -l INFO \
  from-subdir-ic \
    -l INFO \
    -i ./subdir/ \
  add-annotation-overlay-ic \
    -l INFO \
  to-data \
    -l INFO \
    -o ./17flowers-ic-overlay

Here is an example (image_0008.jpg):

Annotation overlays (object detection)

Download the 17 flowers object detection VOC XML dataset and extract it.

With the add-annotation-overlay-od filter you can overlay the objects on the images:

idc-convert -l INFO \
  from-voc-od \
    -l INFO \
    -i "./voc/*.xml" \
  add-annotation-overlay-od \
    -l INFO \
  to-data \
    -l INFO \
    -o ./17flowers-od-overlay

Here is an example (image_0014.jpg):

Annotation overlays (image segmentation)

Download the blue channel archive of the camvid dataset and extract it.

With the add-annotation-overlay-is filter you can overlay the segmentation layers:

idc-convert -l INFO \
  from-blue-channel-is \
    -l INFO \
    -i "./bluechannel/*.png" \
    --labels Animal Archway Bicyclist Bridge Building Car CartLuggagePram Child Column_Pole \
             Fence LaneMkgsDriv LaneMkgsNonDriv Misc_Text MotorcycleScooter OtherMoving ParkingBlock \
             Pedestrian Road RoadShoulder Sidewalk SignSymbol Sky SUVPickupTruck TrafficCone \
             TrafficLight Train Tree Truck_Bus Tunnel VegetationMisc Void Wall \
  add-annotation-overlay-is \
    -l INFO \
  to-data \
    -l INFO \
    -o ./camvid-is-overlay

Here is an example (0001TP_006900.jpg):

Combining all annotations in one image

Using the to-annotation-overlay-od writer, you can generate a single PNG file that contains the outlines (bbox or polygon) of your object detection annotations. That way, you can see whether certain areas are under-represented with annotations and where hotspots are:

idc-convert -l INFO \
  from-voc-od \
    -l INFO \
    -i "./voc/*.xml" \
  to-annotation-overlay-od \
    -l INFO \
    -o ./17flowers-annotations-overlay.png

The 17 flowers dataset looks like this: