| nparray | NPARRAY,IMAGE,MASK | | 2-D or 3-D ndarray to render. image mode supports 1/3/4 channels; normalize/heatmap support 1/2/3/4 channels and draw multi-channel inputs as quadrants. More channels than that need 'reduce_channels'. A batched IMAGE/MASK/LATENT or a 4-D [B,H,W,C] array renders every frame. For other shapes use 'Inspect CV Data' instead. |
| render | COMBO | image | image: show uint8/float values as-is. normalize: min-max stretch to 0-255 (for gradients/disparity/score maps). heatmap: normalize then apply the 'colormap' palette. |
| color_scaleopt | BOOLEAN | false | Draw a vertical colour/value scale bar with min..max labels next to the preview. The labels are the array's value range ('normalize'/'heatmap' map that range to the bar exactly). On a batch each frame gets its own range unless 'range_mode' says otherwise, and the label column is widened to the batch's widest labels so the frames stay the same width (the surplus is black). |
| reduce_channelsopt | COMBO | KEEP_ALL | Collapse a multi-channel array to ONE 2-D map before rendering, instead of drawing the channels as quadrants. KEEP_ALL leaves the array alone. REDUCE_SUM/AVG/MAX/MIN/SUM2 are cv2.reduce's reductions across the channel axis. MAGNITUDE_L2 is sqrt(sum of squares) - use it for SIGNED fields (optical flow, gradients), where AVG and SUM cancel to near zero. This is also the only way to preview an array with more than 4 channels, such as a LATENT. The reduction runs in float32, so SUM cannot wrap. |
| colormapopt | COMBO | COLORMAP_VIRIDIS | Palette for 'heatmap' (ignored by the other render modes). VIRIDIS/INFERNO are perceptually uniform and the right default for data; JET is the classic rainbow and misreads as structure. The colour bar and the NaN marker both follow this choice. |
| range_modeopt | COMBO | per frame (auto) | Which value range 'normalize'/'heatmap' stretch to 0-255 (ignored by 'image'). per frame: each frame uses its own min..max - the default, but it makes the frames of a batch incomparable and a near-static sequence flicker. batch: one min..max shared by every frame. manual: the fixed 'vmin'..'vmax' below, with out-of-range values clamped. |
| vminopt | FLOAT | 0.00 | Bottom of the value range, used only when 'range_mode' is manual. If vmin >= vmax the node falls back to the automatic range rather than failing. |
| vmaxopt | FLOAT | 1.00 | Top of the value range, used only when 'range_mode' is manual. |
| clip_percentopt | FLOAT | 0.00–49 | Robust stretch for the automatic range modes: ignore this percentage of values at EACH end when picking min..max, so a single outlier (a hot pixel, a huge distanceTransform corner) cannot flatten the rest of the map to black. 2.0 is a good starting point. 0 = plain min..max. Ignored when 'range_mode' is manual. |
