SEGS BBox Overlay:
SegsOverlay adds a detection preview to an image so you can see which areas an Impact Pack detector found. It draws an outlined box around each segment and places a filled caption beside it with the class name and, if selected, a confidence percentage. It can also tint the detected masks to show their shapes more clearly. This makes it easier to check detections visually while keeping the original segment data available for later steps, such as a Detailer workflow. If the image has a different resolution from the one recorded with the segments, the node proportionally scales the boxes and masks for the preview.
SEGS BBox Overlay Input Parameters:
image
The IMAGE batch to annotate. Use the image that the detector analyzed when possible, so the boxes and masks line up with the pictured subjects. The node draws the same supplied detections on each frame in the batch. No default or adjustable range is specified for this required input.
segs
The required SEGS data containing the detected segments, typically from an Impact Pack detector such as SEGM Detector (SEGS). Each segment supplies the information needed to draw its box, caption, and mask. The node uses the size recorded in SEGS to scale detections when it differs from the image size, and passes the SEGS data through unchanged. No default or adjustable range is specified.
label_format
Controls what appears in each caption. The documented options are label for the class name alone and label+confidence for the class name followed by its confidence percentage, such as face 91%. The default is label+confidence. Choose the shorter format when you want a less crowded preview.
draw_masks
Determines whether detected masks receive a color tint in addition to the outlined boxes and captions. The default is True (masks on); False gives you boxes without mask tinting. Turn tinting on when the precise detected shape matters, or off when you want a clearer view of the underlying image.
mask_alpha
Sets the strength of the mask tint when draw_masks is enabled. The minimum is 0.0, the maximum is 1.0, the default is 0.45, and the control moves in steps of 0.05. Lower values make the tint less prominent; higher values make it stronger. This setting does not remove the boxes or captions.
thickness
Sets the box border width in pixels. The minimum is 0, the maximum is 32, the default is 0, and the control moves in steps of 1. A value of 0 automatically chooses a width based on image height; choose a positive value when you need a specific border width.
font_scale
Multiplies the automatically chosen caption text size. The minimum is 0.1, the maximum is 10.0, the default is 1.0, and the control moves in steps of 0.1. Increase it if captions are hard to read, or decrease it if they occupy too much of a crowded preview.
color_mode
Chooses how detection colors are assigned. The documented options are auto, which gives each class a stable color from a built-in palette, and single, which uses one chosen color for all detections. The default is auto. Use auto to distinguish classes visually, or single for a consistent annotation color.
color_r
Sets the red channel of the chosen color when color_mode is single. The minimum is 0, the maximum is 255, the default is 0, and the control moves in steps of 1. It works with color_g and color_b to define the color; it does not select a class-specific color in auto mode.
color_g
Sets the green channel of the chosen color when color_mode is single. The minimum is 0, the maximum is 255, the default is 255, and the control moves in steps of 1. Together with the default red and blue values, this specifies green as the default single color.
color_b
Sets the blue channel of the chosen color when color_mode is single. The minimum is 0, the maximum is 255, the default is 0, and the control moves in steps of 1. Adjust it alongside the red and green channels to choose the color you want for all detections.
SEGS BBox Overlay Output Parameters:
image
The annotated IMAGE batch, with detection boxes and captions drawn on its frames and mask tinting applied if enabled. This is the output to connect to an image preview or other step where you want to see the detections. If SEGS contains no detections, the node returns the image frames without drawing annotations.
segs
The original SEGS value, passed through unchanged. Connect this output to a later node that needs the actual segment data, such as a Detailer; the visible boxes and tints on the IMAGE output are a preview, not a replacement for SEGS. Any scaling used to draw the preview does not change this output.
SEGS BBox Overlay Usage Tips:
- Connect the detector’s SEGS result and the corresponding image to inspect what was detected before sending SEGS onward to a Detailer. A corresponding image gives you the clearest way to judge alignment.
- Keep
draw_maskson when checking whether a segment follows a subject’s shape. Turn it off when the tint obscures details you need to inspect; adjustmask_alphato make the tint subtler or stronger. - Start with
thicknessat0for an automatically sized border andfont_scaleat1.0. Change them only if boxes or captions are difficult to read at your image size. - Choose
label+confidencewhen evaluating detection scores. Uselabelwhen the scores make the image too busy. - Use
color_modeset toautoto distinguish classes. For one consistent color, selectsingleand setcolor_r,color_g, andcolor_b. - If you use an image at a different resolution from the one recorded in SEGS, the node scales detections proportionally for drawing. Check the preview to confirm they still align with the image content.
SEGS BBox Overlay Common Errors and Solutions:
Expected a SEGS ((h, w), [SEG, ...]) payload, got ...
- Explanation: The supplied SEGS value does not have the expected structure: an image size as height and width, followed by a collection of segments. The node raises this error while reading the payload.
- Solution: Connect a SEGS output from a compatible Impact Pack detector, such as SEGM Detector (SEGS). Check that another node has not replaced the SEGS value with an image, mask, or other data.
Segment ... does not look like an Impact Pack SEG
- Explanation: At least one item in the SEGS collection is missing a readable bounding box, crop region, or label, or one of those values cannot be converted to the expected form.
- Solution: Use segments produced by a compatible SEGS detector. If another node constructs or modifies the segments, check that each one retains its bounding box, crop region, label, confidence, and cropped mask.
No boxes or masks appear, without an error
- Explanation: An empty SEGS collection is not treated as an error. The node returns the image without drawing detections.
- Solution: Inspect the detector’s results and settings to confirm that it found segments. Also confirm that the intended detector output is connected to
segs.
Boxes or mask tints do not line up with the image
- Explanation: The node proportionally scales detections when the image dimensions differ from the size recorded in SEGS. If the image content has also been cropped, repositioned, or otherwise changed, proportional scaling alone may not restore alignment.
- Solution: Use the image corresponding to the SEGS detections. If you need to transform the image, ensure the segment coordinates and masks are transformed to match before previewing them.
Masks are not visible, but boxes appear
- Explanation: Mask tinting is disabled when
draw_masksisFalse; a very lowmask_alphacan also make the tint hard to see. - Solution: Set
draw_maskstoTrueand increasemask_alphafrom a low value. If a particular segment still has no visible tint, check that its SEGS data includes a usable cropped mask.
