转换ncnn样例提供的squeezenet_v1.1模型，运行结果错误，

[shijie-nv]

squeezenet_v1.1的模型和数据来自ncnn的github，测试图片cat.jpg来自 tengine example

转换 ncnn/examples 里的 squeezenet_v1.1.param convert_tool -f ncnn -p squeezenet_v1.1.param -m squeezenet_v1.1.bin -o squeezenet_v1.1.tmfile 测试结果完全不对 tm_classification -m squeezenet_v1.1_from_ncnn_new.tmfile -i images/cat.jpg -g 227,227 -w 104.0,117.0,123.0 -s 1.0,1.0,1.0 tengine-lite library version: 1.5-dev

model file : squeezenet_v1.1_from_ncnn_new.tmfile image file : images/cat.jpg img_h, img_w, scale[3], mean[3] : 227 227 , 1.000 1.000 1.000, 104.0 117.0 123.0 Repeat 1 times, thread 1, avg time 57.75 ms, max_time 57.75 ms, min_time 57.75 ms

0.000000, -489353152 0.000000, 0 -0.001556, 1047165711 0.000000, 33 0.000000, -489353120

转换ncnn/examples 里的 squeezenet_v1.1.caffemodel convert_tool -f caffe -p squeezenet_v1.1.prototxt -m squeezenet_v1.1.caffemodel -o squeezenet_v1.1.tmfile 测试结果正常 ./tm_classification -m squeezenet_v1.1.tmfile -i images/cat.jpg -g 227,227 -w 104.0,117.0,123.0 -s 1.0,1.0,1.0 tengine-lite library version: 1.5-dev

model file : squeezenet_v1.1.tmfile image file : images/cat.jpg img_h, img_w, scale[3], mean[3] : 227 227 , 1.000 1.000 1.000, 104.0 117.0 123.0 Repeat 1 times, thread 1, avg time 58.01 ms, max_time 58.01 ms, min_time 58.01 ms

0.225130, 281 0.222654, 282 0.167026, 278 0.090635, 285 0.083020, 277

ncnn的运行结果 ./squeezenet images/cat.jpg 281 = 0.227176 282 = 0.221582 278 = 0.162667

我用Netron查看了两个tmfile，结构是一样的。 squeezenet_v1.1_from_caffe.tmfile.zip squeezenet_v1.1_from_ncnn.tmfile.zip