Encapsulate all pixel data and associated metadata. In a streaming call, this field is allowed to be null, which simply is a request to re-analyze the last submitted image with the updated settings. If this is the first call in a stream, the server will not return results, but simply update and remember the settings.

Other parameters associated with the request, such as score threshold etc. In a streaming call, this field is allowed to be null, in which case the server will use the setings of the previous request.

A field_mask to indicate the requested fields in the reply. Default is to return the full response.

Minimal cell area to be consider for detection. The unit is decided by the physical unit of the image, i.e., if the physical pixel size unit is "nm", then the unit here is nm^2 (2D) or nm^3 (3D). However, if the physical pixel size field is not set, then the unit here is pixels.

Minimal detection score in order to be considered. If unset, the default value is chosen server-side.

Threshold for binarize segmentation logits. If unset, default to 0.

A threshold value for remove overlapping detections. If two detection has an IOU > this threshold, the one with the lower score will be eliminated from the detection list. Nothing will be removed if the field is not set.

Specify the model name for servers that support more than one model variants.

An opaque data structure that specify a server side model configuration. An important use case is to perform few-shot model adaptations. The model adaptation call returns a new model configuration. This stucture should be included in all new inference requests, so that the server understand that it should perform computation using the adapted model.

A hint for the typical cell diameter. If the physical pixel size is known, the value takes the same physical unit. Otherwise, the unit is pixel.

A direct suggestion for image scaling. A nonnegative value. However, requesting a very large scaling factor will likely be refused by the server due to the potential out-of-memory problem.

If the center`` has no z field, neither shouldradius`

Optionally one can cache the computed normal vectors here

Optionally one can cache the associated face values here

a list of region of interest

additional text annotation

Encapsulated image data

Optional additional information regarding the image that will be utilized by the RPC call. For example, some SegmentAnything type models may ask the user to provide a tentative bounding box for the objects that the users are interested in.

All vertices of the mesh

a list of op names

Pixel values

A string describing the dimension order of the pixel data, e.g. "XYZCT" The string must be a permutaion of these 5 letters even if some of the dimension doesn't exist in the data.

A string describing the data type using numpy's convention. E.g., 'i8' represent int64. Endianness indicator, e.g. '=i8', is allowed but will be ignored. Instead the endianness will be determined by the field within the Bindata structure.

Dimensional size of pixel data array [units:none]

Physical size of a pixel.

physical size unit, default "µm"

time_increment is used for time series that have a global timing specification instead of per-timepoint timing info. For example in a video stream. Units are set by TimeIncrementUnit.

time unit, default "s"

for large image dataset, it is necessary to send a stream of image patches instead of one monolithic data structure of the whole dataset. The offset values allows one to specify the relative locations of each patch.

Input image for processing

Specify a operator name because the server may implement multiple operations.

A field_mask to indicate the requested fields in the reply. Default is to return the full response. May be ignored by the server.

Output image after processing

Additonal information the server may return as a text string