yolov3的 例子 在 v851s 无法运行

[super0759]

1， 已经 按照文档 ，生成了 一个yolo3-tiny.nb 的 模型 。
2， yolo3代码

/*-------------------------------------------
                      Main Functions
    -------------------------------------------*/
    int main(int argc, char **argv)
    {
        vip_status_e status = VIP_SUCCESS;
        vip_network_items *network_items = VIP_NULL;
    
        if (argc < 3)
        {
            printf("%s\n", usage);
            printf("Arguments count %d is incorrect!\n", argc);
            return -1;
        }
    
        /* Initialize  vip lite */
        status = vip_init(22 * 1024 * 1024);
        _CHECK_STATUS(status, final)；

3， 运行看来 内存够，但是 报这个错误

root@TinaLinux:/mnt/UDISK# ls
Samples            human.nb           network_binary.nb  sample
face.nb            images             overlay            yolo3_tiny.nb
root@TinaLinux:/mnt/UDISK# yolov3 yolo3_tiny.nb images/416.jpg
[0xb6f45560]vip_init[104], [ 5017.823292] npu[3f3][3f3] vipcore, device init..

The version of Viplite is: 1.8[ 5017.830737] set_vip_power_clk ON
.0-0-AW-2022-04-21
[ 5017.839122] enter aw vip mem alloc size 23068672
[ 5017.862983] aw_vip_mem_alloc vir 0xc5801000, phy 0x48800000
[ 5017.869450] npu[3f3][3f3] gckvip_drv_init  kernel logical phy address=0x4880000          0  virtual =0xc5801000
Create Neural Network: 772.79ms or 772792.31us
Start run graph [1] times...
Run the 1 time: 20.64ms or 20636.42us
vip run network execution time:
Total   20.90ms or 20897.33us
Average 20.90ms or 20897.33us
data_format=2 buff_size=43095
data_format=2 buff_size=172380
[ 5018.989366] npu[3f3][3f3] gckvip_drv_exit, aw_vip_mem_free
[ 5018.995563] aw_vip_mem_free vir 0xc5801000, phy 0x48800000
[ 5019.001841] aw_vip_mem_free dma_unmap_sg_atrs
[ 5019.008456] aw_vip_mem_free ion_unmap_kernel
[ 5019.015041] aw_vip_mem_free ion_free
[ 5019.028767] aw_vip_mem_free ion_client_destroy
[ 5019.034848] npu[3f3][3f3] vipcore, device un-init..
Segmentation fault (core dumped)
root@TinaLinux:/mnt/UDISK#

[WhycanService]

@super0759
tiny的输出是2个tensor，而标准v3是三个tensor，可以在后处理部分删除一个tensor的解析。这里Segmentation fault (core dumped)就是因为只输出了两个tensor但是后处理需要三个tensor导致数据错误

[AI379918679]

@whycanservice 如果只有一个output输出呢，大佬，主要改vnn_post_process.cpp这个文件吗？