ComfyUI Node
CV Saliency
Estimates which parts of the image draw the eye (cv2.saliency) - no model file needed. Useful for attention-weighted crops, auto-framing a subject, or steering a mask towards what matters. 'spectral-residual' is a fast global frequency-domain estimate; 'fine-grained' works on local centre-surround contrast and gives sharper object outlines. Outputs a 0-1 saliency MASK plus an auto-thresholded binary mask.
CV Saliency
- image
- saliency
- binary
◄methodfine-grained►
Categoryimage/CV/contrib
Inputs (2)
| Name | Type | Default | Description |
|---|---|---|---|
| image | NPARRAY,IMAGE | Image to analyse (3-channel). Frame 0 of a batch. Accepts a ComfyUI IMAGE/MASK directly (frame 0 of a batch) or an NPARRAY. Arithmetic ops (add, multiply, etc.) process the full IMAGE batch when both inputs have the same batch size. | |
| method | COMBO | fine-grained | 'fine-grained' follows object boundaries more closely; 'spectral-residual' is faster and coarser, highlighting whatever is unusual in the frequency spectrum. |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| saliency | MASK | 0-1 MASK: how salient each pixel is. Feed it to 'CV Color Map' for a heatmap, or use it directly as a soft weight. |
| binary | MASK | Auto-thresholded 0/1 MASK of the salient region (Otsu-style, computed by OpenCV). Empty is a valid result on a flat image. |