Can't build ssd-mobilenet caffe model to tensorrt

[SokPhanith]

Hello @dusty-nv . I train my custom dataset object detection caffe model SSD-Moblienet from https://github.com/chuanqi305/MobileNet-SSD. I test a simple is working for me. but when I try build tensorrt with command line didn't work.

/detectnet-camera --model=/home/phanith/MobileNet-SSD/dataset/output/mobilenet_iter_3500.caffemodel --prototxt=/home/phanith/MobileNet-SSD/dataset/prototxt/MobileNetSSD_deploy.prototxt --labels=/home/phanith/MobileNet-SSD/dataset/list/labels.txt --input-blob="data" --output-bbox="detection_out" --mean-pixel=127.5 csi://0 --input-flip=rotate-180

-Error:

Warning: Flatten layer ignored. TensorRT implicitly flattens input to FullyConnected layers, but in other circumstances this will result in undefined behavior. Warning: Flatten layer ignored. TensorRT implicitly flattens input to FullyConnected layers, but in other circumstances this will result in undefined behavior. Warning: Flatten layer ignored. TensorRT implicitly flattens input to FullyConnected layers, but in other circumstances this will result in undefined behavior. Warning: Flatten layer ignored. TensorRT implicitly flattens input to FullyConnected layers, but in other circumstances this will result in undefined behavior. Warning: Flatten layer ignored. TensorRT implicitly flattens input to FullyConnected layers, but in other circumstances this will result in undefined behavior. Warning: Flatten layer ignored. TensorRT implicitly flattens input to FullyConnected layers, but in other circumstances this will result in undefined behavior. Warning: Flatten layer ignored. TensorRT implicitly flattens input to FullyConnected layers, but in other circumstances this will result in undefined behavior. Warning: Flatten layer ignored. TensorRT implicitly flattens input to FullyConnected layers, but in other circumstances this will result in undefined behavior. Warning: Flatten layer ignored. TensorRT implicitly flattens input to FullyConnected layers, but in other circumstances this will result in undefined behavior. Warning: Flatten layer ignored. TensorRT implicitly flattens input to FullyConnected layers, but in other circumstances this will result in undefined behavior. Warning: Flatten layer ignored. TensorRT implicitly flattens input to FullyConnected layers, but in other circumstances this will result in undefined behavior. Warning: Flatten layer ignored. TensorRT implicitly flattens input to FullyConnected layers, but in other circumstances this will result in undefined behavior. [TRT] mbox_loc: all concat input tensors must have the same dimensions except on the concatenation axis (0), but dimensions mismatched at index 1. Input 0 shape: [19,19,12], Input 1 shape: [10,10,24] [TRT] mbox_loc: all concat input tensors must have the same dimensions except on the concatenation axis (0), but dimensions mismatched at index 1. Input 0 shape: [19,19,12], Input 1 shape: [10,10,24] Caffe Parser: Invalid axis in softmax layer - TensorRT expects NCHW input. Negative axis is not supported in TensorRT, please use positive axis indexing error parsing layer type Softmax index 186 [TRT] device GPU, failed to parse caffe network [TRT] device GPU, failed to load /home/phanith/MobileNet-SSD/dataset/output/mobilenet_iter_3500.caffemodel [TRT] detectNet -- failed to initialize. detectnet: failed to load detectNet model

[dusty-nv]

Hi @SokPhanith, I haven't tried SSD-Mobilenet through Caffe before. I recommend to use this PyTorch version here, which is known to be working: https://github.com/dusty-nv/jetson-inference/blob/master/docs/pytorch-ssd.md

[SokPhanith]

I used to Re-train on it https://github.com/dusty-nv/jetson-inference/blob/master/docs/pytorch-ssd.md but I saw /detectnet-camera support with caffe model too. Can you give me Github source or caffe model that SSD-Mobilenet can build with /detectnet-camera

[dusty-nv]

That is only the older 'detectNet' DNN architecture trained with DIGITS that worked with caffe. The newer SSD-Mobilenet model that works with TensorRT is trained with PyTorch. I don't know of SSD-Mobilenet from Caffe that works with TensorRT.