It is so easy to train vision models with the UI. You can prototype a model in minutes to validate ideas and, if it works, you can refine it to production-ready on the same platform. It's easy to switch models, but that is not needed with Roboflow's RF-DETR since it performs so well. Review collected by and hosted on G2.com.

Sometime the annotation UI is a little buggy but a refresh always fixes the issue. Review collected by and hosted on G2.com.

It has been extremely easy to access various computer vision strategies without having to learn the intricacies of each. For example, I was able to train a keypoint detection model and then run it against a video of vehicles in under a day. Normally there'd be debugging for days trying to find where code has gone wrong, but they have a plug-n-play option with "Workflows" to use established code chunks and keep rolling forward. And then on top of that, they provide a way to access that same workflow via Python and expand on it as needed for your custom use-case. Review collected by and hosted on G2.com.

The downside is that there are so many ways to apply computer vision, sometimes you stumble on one that they have not formally implemented so then you have to go the manual route. Kind of the case with any complex software platform like this. Although they have also been receptive in taking feedback from individuals for feature requests and so the service is expanding daily. But overall, they seem to have what I need 99% of the time so I always go there first and usually get what I'm looking for. Review collected by and hosted on G2.com.

We used several datasets in our recent paper on out-of-stock detection from a robotic platform [Deep learning based empty shelf detection based on autonomous mobile robot https://www.sciencedirect.com/science/article/pii/S1077314226000640].

They were very easy to find and use. Review collected by and hosted on G2.com.

More persons should upload their data :-) Review collected by and hosted on G2.com.

I really liked how easy it was to upload and annotate images. The interface is clean and intuitive, and the data augmentation options are very helpful for improving model performance. Review collected by and hosted on G2.com.

The only thing I didn’t like is that some advanced features require a paid plan, but the free version is still very useful. Review collected by and hosted on G2.com.

Roboflow offers a comprehensive end-to-end platform for computer vision projects. Its annotation tools are both intuitive and efficient, allowing for quick and organized dataset preparation. The model training workflow is seamless and well-optimized, and we consistently achieve highly accurate results with models trained using Roboflow.

One of the most impressive aspects is Roboflow’s deployment and model utilization features, which provide robust monitoring and performance tracking. These capabilities simplify the integration of models into production workflows and make it easy to monitor their effectiveness in real-world scenarios.

In summary, Roboflow is a dependable, user-friendly, and production-ready platform that greatly streamlines the entire machine learning lifecycle, from data labeling through to deployment. I highly recommend it for teams engaged in computer vision projects.

The customer support is great, had an overall good experience with the team. Review collected by and hosted on G2.com.

Roboflow is a strong platform overall, but there is room for improvement, particularly in allowing users to train and deploy their own advanced models—such as SAM 2.1 or other cutting-edge architectures—directly within the platform.

While Roboflow already offers features like auto-annotation and smart labeling, giving users the ability to train or fine-tune large foundation models (for example, Segment Anything, Grounding DINO, or similar) on their own datasets would significantly increase the platform’s capabilities. This would enable users to produce high-quality annotations with less manual work, while also retaining full control over how their models are customized and how accurate they are.

It would also be beneficial if Roboflow included a wider selection of pretrained models that could be used immediately for annotation and deployment. This would boost productivity and make it easier for teams to take advantage of GPU acceleration and model versioning, all without needing to turn to outside tools.

Finally, expanding the platform’s support for fine-tuning workflows—beyond what is currently available with Roboflow Train templates—would be a valuable enhancement. Review collected by and hosted on G2.com.

I've done a little bit of computer vision work in the past, but it's not my primary skillset. Even trained a vision model or two. There's so much that goes into building and (importantly) deploying vision models. Roboflow is great in that it takes all the steps involved and puts them into one platform. It's incredibly easy to get started with.

I can easily send images to roboflow and then annotate my dataset. The annotation tool itself is valuable, it's pretty easy/intuitive. I really like how easy it is to divvy work up between colleagues on the labeling side. It's great having a model that you can continuously monitor and improve, without being an ML expert.

Roboflow doesn't expect you to be an expert to get expert level computer vision models trained. You can pretty much start from a very high level understanding of vision models and in a few days have a finely tuned model for your specific use case.

The team is incredibly responsive and helpful when you need help. They also build some popular and useful open source projects (like the supervision project).

We've used roboflow for multiple various projects over the years and it's become our go to starting point for any vision based project we do.

We were able to build a computer vision model that is assisting us in getting much higher utilization of our raw metal on our laser cutting machine with just a few days of engineering effort.

I've happily recommended roboflow to other engineers and startup founders in the past and will keep doing so. Review collected by and hosted on G2.com.

Auto label feature looks promising but I could only use it for bounding box. I'd like to use an auto labeler for instance segmentation.

Tiny quality of life thing for labeling: would be nice to have a keyboard shortcut on the left hand side of the keyboard for completing an annotation. Right now you have to press Enter, so you either have to let go of your mouse, or move your left hand over to the right side of the keyboard. Review collected by and hosted on G2.com.

Roboflow is the clear choice for computer vision at scale. After trying Cognex, Keyence, and other point tools we moved to Roboflow because it lets us roll out every vision application across multiple plants from one cloud dashboard and push to edge devices in minutes. The platform works with inexpensive off the shelf cameras and consumer GPUs, cutting hardware spend by more than half. New features drop weekly and every new model such as YOLO or SAM appears on launch day so we can benchmark the upgrade on the same dataset with a few clicks. Unified annotation, training, version control, and inference keep our workflow tight and fast. Support is outstanding; Roboflow engineers jump on calls, share code, and really listen. High trust, rapid ROI, and a product roadmap that keeps delivering make Roboflow the best engine for quality inspection, safety, and predictive maintenance in modern manufacturing.

It's very easy to deploy with multiple deployment options whether it's in the cloud for a pilot, a Nvidia Jetson for an individual line or server based for a full factory solution. Their device monitoring is great for tracking an entire plant and receiving alerts for defect trends or downtime. I like that you can use the GUI workflows or fully deploy with Python all using the same models. I have moved to always recommending Roboflow for their full end to end integrated solution. Review collected by and hosted on G2.com.

I don't love subscriptions, but I guess this is just the world we live in today. They do offer flexibility in their pricing to shift the cost to CapEx Review collected by and hosted on G2.com.

Cloud based, easy to share, collaborate, and maintain the project with your team members. Outsourcing support Review collected by and hosted on G2.com.

Sometimes I run into platform bugs—for example, my batches disappear when I navigate around, then show up again after a while. Review collected by and hosted on G2.com.

The interface is user-friendly and requires little to no experience for beginners. Multiple tutorials are available online in case anyone needs it.

The auto-annotating feature was most helpful in generating my dataset and saved me a lot of time.

I also used the software as a backend editor for misclassified instance segmentation images. Importing, editing, and exporting were all very smooth. Best annotation tool out there. Review collected by and hosted on G2.com.

For most object instances with a gap or hole inside, the auto-annotator was not able to separate just the instance from the background. It was a little tedious in some cases. Review collected by and hosted on G2.com.

The model metrics are fantastic alongside the diagrams. Review collected by and hosted on G2.com.

I think the pricing structure is a little confusing and could be more clear. Review collected by and hosted on G2.com.