One of my favorite tools for people to dive into deep learning is Darknet. Not this one. This is a neural network framework in c. But despite that, this is nice in a way that it can function as an application.
Before you use it, you need the weights file, it is linked in the repository. The command is,
darknet detector test <labels> <config> <weights> <image path>. Example of usage is the following.
./darknet detector test cfg/coco.data cfg/yolov4-tiny.cfg yolov4-tiny.weights data/person.jpg
By default it recognize 80 objects listed here. It works with image and video.
While it can function as a commandline application, it is also a research tools. You can also train with your own dataset. An example of it is at my project, which is not about kuih anymore.
In training darknet save the weight in a weight file, which is a binary. If you want to deploy to tflite, you may use this tool. The converstion tools also have example to use for android app. Still trying to figure out deployment to browser.