What Viator Florence Actually Is
Viator Florence is a specialized tool used primarily in medical and clinical research settings for processing imaging data from the Florence protocol. It isn't a consumer product. The software handles batch processing, image segmentation, and basic quantitative analysis of datasets generated under standardized protocols. People come across it through academic partnerships or hospital IT departments. It's not available as a general download from a storefront. The installation process is straightforward if you already have the prerequisite environment. You need Python 3.9 or 3.10, the standard scientific stack (NumPy, SciPy, scikit-image), and then the Viator Florence package itself. The GitHub repository is usually pinned to a specific commit on the project's internal staging server. Clone it, run the install script with the requirements file, and configure the YAML config to point at your data directory. That's roughly it for setup. The workflow runs like this: point the config at your input directory, set your output path, define which preprocessing pipeline you want (most people use the default), and run the main script. Processing time depends on your hardware and dataset size. On a typical workstations with an SSD and 32GB RAM, a batch of about 200 scans takes roughly 40 to 90 minutes. Larger batches scale linearly unless you hit memory limits, in which case the script will error out around 150GB of intermediate data.
I ran into a specific issue last year with a dataset where the DICOM headers had inconsistent patient ID formatting. Viator Florence expects a strict 8-character alphanumeric format in the PatientID field. About a third of the files in that particular study had leading zeros stripped or mixed-case variations, and the pipeline would silently skip them without any warning in the log. The workaround was a quick Python script using pydicom to normalize all IDs before feeding them into the pipeline. It took me maybe 15 minutes to write and saved me from discovering the problem after the batch finished and realizing half the data was gone.
Counter-intuitive things nobody mentions
First, the built-in quality control flags are not reliable on their own. They use simple threshold-based checks that miss subtle artifacts like motion blur in the mid-range of severity. I learned this the hard way when a batch came back flagged as all-pass and I later spotted consistent motion artifacts in about 12 percent of the volumes during a visual review. The fix is to always run a secondary visual check on a random 10 percent sample before trusting the QC pass rate. Second, the default memory allocation assumes you're running on a dedicated machine. If you're on a shared cluster node with concurrent users, the default settings will compete for RAM and cause swap thrashing, which doubles your processing time. Setting the memory cap parameter in the config to about 70 percent of your node's total RAM fixes this. The documentation mentions it in a single line near the bottom of the config reference.
Get the Full Details
Known limitations and when to look elsewhere
Viator Florence doesn't support GPU acceleration. If you're processing thousands of scans regularly, this is a real bottleneck. It also only handles one imaging modality well out of the box. Second modality support requires writing custom preprocessing modules, which the developers document but don't provide templates for. If your use case involves multi-center studies with highly heterogeneous scanner manufacturers, you'll spend more time normalizing data to make Viator Florence tolerate it than you would just using a different pipeline designed for that from the start. In those cases, tools like fMRIPrep for functional data or NiftySeg for segmentation tasks tend to handle heterogeneity better with less manual intervention. There's no official GUI either. Everything is command-line driven, which is fine if you're comfortable in a terminal and write your own automation wrappers. It's a friction point for collaborators who aren't comfortable with that workflow.
The project is actively maintained but updates are infrequent. Major releases happen maybe once or twice a year. Bug fixes land faster but sometimes break backward compatibility with older config formats, so keep your configs version-tagged if you plan to keep datasets around long-term.
Where to get it
The current stable version is hosted on the project's internal repository, which requires institutional authentication to access. Check with the lead researcher or your hospital's IT department for the login. There is no public PyPI package. Beta versions occasionally appear on a separate staging URL that the team shares via their mailing list. If you found a third-party site offering a standalone download, it's not authorized and could contain modified code.
