siryin-eason / efficientnet-bifpn-bilstm-behavior-recognition

An efficient and multi-scale feature fusion behavior recognition algorithm

EfficientNet-BiFPN-BiLSTM (Behavior recognition)

An efficient and multi-scale feature fusion behavior recognition algorithm

Environment:

python>=3.6

pytorch>=1.2

The overall frame diagram is as follows:

Datasets:

Following the format of UCF101 action recognition.

Run steps:

1. Modify the "dict_data" of readpkl.py to your own category to generate your own label.pkl for training.
2. Run train.py, remember to modify the data set address.
3. During operation, all loss npy files will be saved for visualization, and all models will be saved under weights.
4. Run train.py, remember to modify the test data set address and the name of your trained model. In addition, Ranger and BiFPN modules included in this project could also be used for other computer vision tasks.

efficientnet-bifpn-bilstm-behavior-recognition's Issues

代码、文章请教

你好，我对这个非常感兴趣。我想问一下，GitHub上代码是完整的吗？？我应该在哪找到你所发的对应的文章呢？？还望不吝赐教，非常感谢!

question

def init_weights(self): for m in self.modules(): if isinstance(m, nn.Conv2d): xavier_init(m, distribution='uniform')

请问，xavier_init是什么意思，这里报错了，谢谢