Shape the future of FME with your ideas
Open ideas have been reviewed by our Customer Success team and are open for commenting and voting.
Allow for the python caller to use Python Virtual EnvironmentsComms around it were had a couple of years ago, but nothing seems to have progressed - https://community.safe.com/s/question/0D54Q000080hOlhSAE/is-there-a-way-to-use-a-python-virtual-environment
I work with time series data a lot, and to keep some slack with regards to “race conditions” across my own flows and other upstream jobs, I tend to use rather generous time windows and “full rerun at the end of the day” configurations to ensure I don’t leave any time gaps in the data destination. That means a lot of the features being inserted at the end will have already been inserted by a previous run and will raise a conflict error if there’s some sort of primary key constraint (or create duplicate entries if these constraints are not enforced).This sort of scenario is what UPSERT is for, but UPSERT causes existing rows to be updated by design. The problem is that several databases, such as Postgres and Snowflake don’t handle large amounts of rows being updated frequently very well, since the “updated row” is actually just a duplicate of the old row with its fields updated, while the “replaced row” is just hidden and actually deleted later by a separate process.Given I know any rows already present in the DB would be identical to the ones I’m inserting, I would much rather skip inserting any conflicting rows instead of causing unneeded churn by “updating” them with the same data. Standard SQL handles this with MERGE INTO … WHEN NOT MATCHED THEN INSERT, but there are also non-standard ways like Postgres’ INSERT INTO … ON CONFLICT DO NOTHING. The only workaround I have at this time is inserting all my features into a temporary table and using an SQLExecutor to write that data to its final destination (something not helped by not being able to use actual temporary tables in FME). Being able to set the FeatureWriter to write new features and ignore conflicts from the get go would be a lot more conveinient for me.Related question/community post:
May be terribly niche, but I’ve just come across a situation where having a ‘counter’ that assigns sequential letters in alphabetical order starting at C through to J would have been useful. If it’s of use to anyone else, maybe the Counter could count letters...or gets a cousin called the Alphabetiser?
It would be great to have a direct integration in FME Form to do git commits, etc. in the GUI itself. It could be in the form of a developer command line/powershell or even as a built-in git integration much like what we have with FME Flow. As it stands today, I do this indirectly through Visual Studio. I essentially save my .fmw files in a Visual Studio project, and do all the version control through there. I would love it if I could bypass that altogether.
With libraries like DuckDB in your Python environment, as is the case where you either have ArcGIS Pro or Enterprise on your FME machine, or you’re using ArcGIS Data Interoperability, then it is easy to retrieve potentially millions of features from an S3-API compliant object store or other web source in seconds, but it’s a bottleneck to send them on into the workspace one by one.I would like to see performance like reading CSV files brought to PythonCreator.There is a possibly related existing idea Introduce a Python Dataframe Creator/Transformer but I don’t want to confound using dataframes with this idea, which is fundamentally about performance.Dataframes might be how this idea is implemented, but my guess is that would be a heavy lift for Safe, another way might be a way to output an aggregate feature.
Supports the latest 3dtiles 1.1, supports both single-sided and double-sided models, supports with and without lighting, supports scaling of small models, supports models not deforming when coordinates are large, and supports adding texture images to models in PythonCaller,
Hi,As pointed at in : Please add a writer for on-disk mode.Thanks!
To understand the purpose of the workbench, we will have a look at the name of the workbench at first step. Secondly, we have to go through entire workbench from reader to writer to understand more and understand that all the changes are included or missed out.On the other side, SRS documents will be in another document (i.e., name.docx) file which we have to refer for better clarity.I would be better if workbench have minimum functionality of MSWord with allowing images from SRS documents to refer. This will help us to maintain the same vocabulary between SRS and Workbench. It will even help team members to coordinate in much better approach.It is advisable to place the document just beside Main tab as shown in screenshot. Thank you
The SpatialRelator has the option to output the cardinality of the intersections. It would be really useful to also have the optional functionality to output the actual geometry of the intersection.
Its not obvious but it would be nice to make it easier to see.Without clicking on the value cell, it looks like a single value parameter. When I click on the cell, I see both the drop down and the ellipsis menu. Since there is a default value for MIME_TYPE, user would just populate the drop down and can easily ignore the mime-type. A further enhancement to this would be to make mime_type configured to an Attribute Value choice similar to “value”. If this is available, then I can add a MimeTypeFinder transformer in my workflow to populate the Mime-type parameter of the Multipart Upload.The use case for this would be when using the AddAttachments REST API and a feature with multiple attachments of different mime types supported by the destination format, eg, AGOL and Portal format writers.
You can download subfolders of /Logs but not the entire /Logs folder.
Make the Snapper snap in 3D.
Hi Will FME FORM is getting an object detection functionality.Can FME Users will be able to train a machine to recognise objects of interest within images supplied to it. We see great potential for this machine learning capability in areas of asset management and surveying and train a model by labeling . Thanks
Hi FME InnovatorsI was wondering do we have any ideas/thoughts in future to implement/release any FME transformers and tools in AI Era, Object Detection is evolving alot and Instead of using External API , FME Owned transformer and Labelling.The Era of AI is evolving, I have kept few notes of Object detection for which i was exploring towards it and I was thinking of FME will have its own API , tools and methods in labelling(Training a model within FME) rather than depending on external Api connectors.What is Object Detection?Object Detection is a method in computer vision that detects and identifies objects in an image or video. While image classification predicts a single label for an entire image, object detection finds several objects in a single image, giving each of them a bounding box and a class label.Object detection takes care of two main functions:Localisation Localization - Where is the object? Classification - What is the object? Traditional Machine Learning for Object DetectionBefore the emergence of deep learning, the traditional approach to object detection was handcrafted features and classical ML algorithms. Traditional object detection techniques require you to do manual feature extraction and suffer from problems with too much variation, such as lighting changes, scale changes, and background changes. Haar Cascades Introduced by Viola and Jones (2001). Utilised for initial face detection (e.g. OpenCV’s face detector). Based on Haar-like features and a cascade of classifiers. Histogram of Oriented Gradients (HOG) + SVM Detect objects by utilising gradient orientations. Popularised by Dalal and Triggs for pedestrian detection. More compact and robust than Haar, but computationally expensive. Selective search + SVM Provides region proposals, which are classified. Helped bridge the gap between traditional machine learning and deep learning. While these machine learning methods set the groundwork, they simply could not outpace both the accuracy and scale of the now deep learning models.Deep Learning for Object DetectionDeep learning has transformed object detection by automating feature extraction via Convolutional Neural Networks (CNNs). Deep learning models automatically learn progressively abstract features from the data, improving speed and accuracy.Two-Stage DetectorsTwo-stage detectors separate the region proposal from classification. R-CNN (Regions with CNN Features) Uses Selective Search to propose region proposals. Uses a CNN to extract features from each proposed region and classify each region. Very accurate, but slow (each region is processed independently). Fast R-CNN This model shared convolutional computation across the image plane. It adds an ROI pooling layer to extract features using shared feature maps. Faster than R-CNN, but still not real-time, close to real-time. Faster R-CNN Introduces a Region Proposal Network for end-to-end training and prediction. Achieves accuracies very close to real-time performance. Single-Stage DetectorsSingle-stage detectors eliminate the need for region proposal and are capable of predicting bounding boxes and class labels directly. YOLO (You Only Look Once) This system is targeted for real-time detection. YOLO divides images into a grid and makes predictions about bounding boxes for each cell in the grid. The versions began with YOLOv3, then to YOLOv4, YOLOv5, and continue to the latest - YOLOv8 (the most recent versions now leverage Transformer-based modifications). SSD (Single Shot MultiBox Detector) SSD uses feature maps from multiple convolutional layers to perform detection. SSD offers a good tradeoff between speed and accuracy. RetinaNet RetinaNet introduced 'Focal Loss', or re-weighted losses, to aid in addressing the issue of class imbalance during training. RetinaNet shows good results across a range of benchmarks.Innovative Architectures and Trends (2025)Modern architectures combine CNNs, Transformers, and self-supervised learning techniques for better generalisation. DETR (Detection Transformer) An end-to-end object detection pipeline that employs Transformers. Negates the need for anchor boxes and Non-Max Suppression (NMS). Very accurate but less computationally efficient than YOLO. Vision Transformers (ViT) Attention mechanism (global feature extraction). Used with a hybrid CNN backbone for efficiency. Self-supervised learning (SSL) Models that are pretrained on unlabeled data (MAE, SimCLR) will transfer better with limited labelled datasets. Tools and FrameworksHere are some popular frameworks for implementing object detection: TensorFlow Object Detection API PyTorch + TorchVision Ultralytics YOLOv8 Detectron2 (by Meta AI) MMDetection Thanks
Often, I’m dealing with large datasets and I want to do a quick comparison of the schemas. I only want one record from each. I appreciate that you can set Schema Scanner or Sampler to only sample one record, but it still processes the remaining features to an output port, which is wasting resource in many cases. How about a setting in both made available, that just takes the first (specified) number of features and then stops reading from the source?
If you select a view called “banana” in the Revit reader but the view doesn’t exist in the file (for instance, in a parameterised or generic workspace), the reader will default to the first available view. This can lead to unwanted data being read, and cannot be controlled by the user without accessing the log file.Please improve the Revit reader by:Not reading any data if the selected view is not available Adding the view that was used to access data to the feature attributes
When I have a job that runs for 5-15 minutes (and this is with a subset of data) it would be nice to be able to sneak a look at the cached features at any point during the running workbench. I could click on the green loading box and see what features have loaded so far. This way I can be QC-ing output before the workbench has even completed running.
When a workspace is stopped by clicking the "Stop" button, no feature caches are preserved. It would be great to have an option to preserve these partially built feature caches if a workspace is manually stopped.Usage scenario: I'm running a workspace that takes a long time to run and I notice an issue (error or warning) that I'd like to investigate but don't want to have to wait for the workspace run to complete just to take a look at the problem.
Currently the FME Flow logfile cleanup is a one-size fits all setting. So all job logs older than the interval that’s set get deleted.I would like to propose a more fine-grained control, so on a per-workspace level be able to override that setting. Let’s say for example we have it set to 7 days, but have some jobs that run monthly. It could be beneficial to be able to see the log from the previous run if the current run fails. And on the other hand, if we have jobs that run many (say hundreds), I see limited value in being able to see one from a week ago, so I may want to shorten that retention period to say 1-2 days.
When you add a "file" manual key to an automation which get run by an automation app, the option is to choose a file already in the resources of FME Flow. It would be nice to have an option for a user to upload a file like what's already available in workspace apps. For a User Parameter "File" the user is able to upload a file which gets used by the workspace the workspace app runs. The same would be ideal for automation apps.
Lots of organisations have several FME Flow Instances, Dev, Test, Prod for example.It would be helpful to have the ability to select different colour pallets for each environment to quickly differentiate between each env.For example, you would be able to change the sidebar colour to blue for dev, green for test and red for prod
It would be nice to be able to give each FME instance an individual name (and possible icon) to be able to distinguish different instances without checking the sometimes cryptic or long url. Handling different stages in the same browser sometimes leads to tab change after tab change to get the right instance.Different Instances of FME FlowIt would be possible to directly chose the correct instance if the admin defined name could be shown instead of FME Flow as title and possibly a selected favicon - that is already possible to change going into the webapp but not documented and the path may vary between versions.
ESRI only makes arcSDE connections available via a ‘sde file’ which is a propierity file stored on disk, sometimes C drive, other times on network storage. Often we struggle with access to these files, as some are open and read only access, others have higher privileges to write to the SDE geodatabase.So its confusing to me having *.sde files as “Database Connections”In 2025, we now have the option to store database connection in FME Flow. We can change the dataset path from local to the Shared Resources on FME, either the Engine or the Data folder.Flow database connection with SDE typeThis enables further re-use of the connection within Flow… however the problem then becomes how can FME authors manage the connection in Form?FME Flow connection storage is great, but not necessarily for SDE file database connectionsIn practice, you go to re-use the flow connection for SDE and despite the path being to Resouces (engine and shared to roles) the error is repeatable Error’s connecting to feature types in ArcSDE geodb encountered in 2025.1The workaround for now is to follow Option 2 in article https://support.safe.com/hc/en-us/articles/30212601575693-How-to-Create-and-Manage-Esri-Geodatabase-ArcSDE-Connections-in-FME storing a single SDE file in a network share location that is accessible to both Form and Flow Can Safe software please add an enhancement to help find a better solution surrounding SDE connections and FME?Maybe ESRI have ideas or community wishes to move away from only have *.sde files as the single means to connect to Spatial Database Engine. What's needed is another means/protocol to properly “direct connect” to the DBMS and to include the sde registry.Connect to the DBMS registered for SDE, add a new option to get license and work as SDEA requirement for FME Flow instance is to have ArcServer installed for the licensing of SDE/FGDB. Perhaps this opens up something new.
In the older Transformer Guides, we sadly lost FME Lizard… but it had an overview, components of a simple workspace, basic transformer placement, basic running tips and inspectionIn the latest Transformer Guide we have the new transformers being added but only the transformer description textI’m asking for the graphics for each transformer to be returned. They still exist in the Help documentation, but the latest transformer guide, it lacks that nice visual, simple guide to what a transformer handles:Please rewrite the PDF to include the graphic/illustrations. Surely an FME workspace can transform the help from the web into a PDF and I believe 2025.1 has a nice PDF styler transformer to assist
All the grids in FME servers are not sortable by users, sorting is the very basic function expected by all users from any software these days.Please allow users to click on any column header to sort ascending/descending/removing sort for a given column. The user should be able to hold the shift or control key to choose multiple columns to sort. The sort function should work for all supported data types in the column, e.g. Number, text, date, user, etc.
In-depth courses for all FME skill levels.
Explore our extensive collection of how-to guides and tutorials.
Log an issue with us. Our support team is ready to help.
