刘桂玉
marvin.hpp(1804): error: identifier "CUBLAS_DATA_HALF" is undefined marvin.hpp(3373): error: argument of type "std::size_t" is incompatible with parameter of type "cudnnTensorFormat_t" marvin.hpp(3374): error: argument of type "int *" is incompatible with parameter...
please upload ~
你好你的问题解决了吗，我遇到了类似的问题，网络结构是mask+lstm+TimeDistributed(Dense）， lstm层之后还是有输出的，但是TimeDistributed(Dense）之后的输入就全为0，求大神解答，结构如下 ``` input=Input(shape=(max_len,feat_dim),name='input_layer') input_layer=input mask=Masking(mask_value=0,name='mask_layer')(input_layer) lstm_dt=LSTM(200,return_sequences=True,name='lstm_dt')(mask) fc_dt=TimeDistributed(Dense(3,activation='relu'),name='fc_dt')(lstm_dt) ```
already compile and install tensorrt plugin successfully make install [100%] Built target tensorrt_ops Install the project... -- Install configuration: "" -- Up-to-date: /home/gyl/projects/BEVFormer_tensorrt/TensorRT/lib/libtensorrt_ops.so
tensorrt 8.5.1.7 pypi_0 pypi cuda 11.6 torch 1.12.1+cu116 pypi_0 pypi
hello do you solve the problem ？
这貌似是库版本的问题？请问如何解决？
no, I guess you can try gsplat to optimize the code
I am wondering how to get the semantic labels, since you network only had instance prediction. but you get panoptic result on table 4
can you explain how to get semantic mask for the rendering novel view after getting instance mask from the network?