Lz
> 训练模型后使用python调用摄像头进行跟踪，发现关键点抖动很明显，有什么方法可以解决？ You can refer to this simple implement , https://github.com/610265158/Peppa_Pig_Face_Engine, checking the demo to see if it is satisfied.
Yes, it is easy to add a subnet in the head to do the classification. And it will be more convenient to do tracking with the face landmark model. However,...
Sorry for reply late. We need a better face detector.
Why do you want to run the tflite model on gpu?
Tflite is not fast, bro. It is an engineering problem, please modify the model and then try MNN or NCNN.
The model structure is not designed so fast. The best way is to tune the model structure based on your device and your inference frame work. If you want to...
Hi, Sorry，I can not give you some advices accurately for this question, because i really don't know. As you see, it is a small model, and the 300w dataset is...
Hi, the 300W, and 300VW, actually, is a poor dataset for face landmark task. In my mind, the best way to solve this , is to sample more data contains...
Hi, The model was trained with only the public dataset. However, the dataset is very poor, very hard augmentation was done here to fit the model, which means the model...
抱歉，没有试过。