Hey ProductHunt ✌🏻 Ever wondered if there's a better way to visualize and discuss neural network results apart from Jupyter Notebooks or using cv2? 🔮 Portal is the easiest way to visualize neural network inferences. For ML engineers and data teams, sharing visual results is challenging and time-consuming as people would run cv2 or matplotlib over and over in Jupyter Notebooks just to visualize the results 😿 🕸 So we made a web-based application to do this! This way, users can adjust confidence thresholds, filter tags, change IoUs and visually inspect their models easily! This works on both bounding boxes, masks and videos. 📺 It works on videos so you can look at inference results frame-by-frame while adjusting thresholds, and IoU settings without having to re-run the entire prediction! 🤖 Users can load custom fine-tuned / transfer-learning FasterRCNN, YoloV3, MaskRCNN, MobileNet, and more into Portal! 🧩 For custom architecture, users can easily change the codes in the repository to make it work for your own models! That's why we open-sourced Portal. It's a team sport in OSS and everyone can build their own extensions! 🎁 Features work out of the box without coding! Feel free to ship it along your models or make adjustments to logo to fit your clients' needs. The code are yours to work with and we just want to make complex features accessible to everyone 🙌 Thank you for checking us out and all the love we received in our previous launch at https://www.producthunt.com/post... - your suggestions shaped this new app - so this launch is for all of you ✌🏻

I have just tried using Portal to perform some inferences for one of my object-detection projects and it worked great! All I had to do was register my TensorFlow model, add some images, load the model, and leave the rest to Portal - all without writing any code 👍🏼 I was even able to filter my predictions on a video while the video is playing, which is a huge plus, saving me a lot of time in the process. Looking forward to future updates and developments!

I was part of Portal's pre-release testing and it really blew me away. As an avid Computer Vision enthusiast, I have always found it a hassle to write code from scratch to visualise predictions every time I explore a new model. Being able to seamlessly load models from different architectures and play with the predictions is such a convenience and time-saver! One bonus I've noticed is the ability to navigate freely between video frames without any stutter. Performance is also blazing fast even when using only the CPU. Portal works perfectly on Windows and Linux, and it even works on WSL2 if you have GUI set up (like XServer). I'm definitely adding Portal to my toolbelt in future projects, and I can't wait to see how Portal will evolve over time!

The program ran smoothly for me. Coming from someone who understands the amount of code needed to be written for a single machine learning project, this can be considered as a potential game changer in the industry. I’ll continue to watch the development of this product, and since it is open source I may also contribute to its development.