Xingyi Yang
Xingyi Yang
What would you do on Darknet53
That's pretty easy. The structure of darknet53 is written in yolo3/model.py. you can do whatever you want on the model.
The weight initialization is the key to deep learning. You can initialize the model with different coefficient distribution. To make the model better, pre-train on imagenet is a much powerful...
You can not use the existing model. Pre-trained model means trained in advance for a certain structure. You have to train your own if you need
Here is a good example. https://stackoverflow.com/questions/48457124/where-is-the-kernel-weight-initialization-in-my-cnn-model In keras, it helps you to initialize your weight. You can also choose the specific way of initialization for your experiment.
So what is the exact input size in Pascal VOC evaluation?
 This might be a clearer derivation.
> I trained a ClassicMLP model on MNIST with an input size of 784, hidden layer sizes of [128, 64], and an output size of 10. This model was trained...
Second stage, when training the LDM.
do lr decay, may help