徐静

RNN_convLSTM structure Error

1

``` # RNN_convLSTM structure def RNN_convLSTM(x ): #(1,19,64,64,1) #num_filter, kernei_size x = tf.contrib.slim.conv2d(x, 1, 1, padding='SAME',activation_fn=tf.nn.relu) x = tf.contrib.slim.conv2d(x, 8, 1, padding='SAME', activation_fn=tf.nn.relu) x = tf.contrib.slim.conv2d(x, 16, 1, padding='SAME', activation_fn=tf.nn.relu)...

install error!

1

``` git clone https://github.com/Slyne/ctc_decoder.git && cd ctc_decoder/swig && bash setup.sh ``` ``` python3 -c "import swig_decoders" ``` ModuleNotFoundError: No module named '_swig_decoders'

版本：20240102 量化后table中出现大量的inf: 

How to use the pretrained model to train my own data, when the class number is not 80(coco)?

2

How can I use pretraied model to train my own own data, my class number is not 80 like coco, but my own data only have 3 classess. When I...

If my Annotations have no **segmentation**, Can I train **ExtremeNet** and obtain the **predict Box** and **octagon mask**？

TensorRT

1

模型转成ONNX后再TensorRT中调用出现： `ERROR: Parameter check failed at: ../builder/Network.cpp::addPoolingNd::690, condition: allDimsGtEq(windowSize, 1) && volume(windowSize) < MAX_KERNEL_DIMS_PRODUCT` 的错误。请问在TensorRT下你们是否测试过。非常感谢！！！

关于模型转换的问题

6

大佬问一下，mediapipe中的`face_landmark_with_attention`这个模型你是怎样转换到ncnn的，给个思路，我用了tflite和tfjs两种方式转换出来的和你的不一致。