# Fail to load densebox

**URL:** <https://discuss.pynq.io/t/fail-to-load-densebox/2200>\
**Category:** Support\
**Created:** [January 4, 2021, 2:48am UTC](https://discuss.pynq.io/t/fail-to-load-densebox/2200 "2021-01-04T02:48:04Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![pkchen](https://sea1.discourse-cdn.com/flex019/user_avatar/discuss.pynq.io/pkchen/32/1635_2.png) [@pkchen](https://discuss.pynq.io/u/pkchen)\
**Post date:** [January 4, 2021, 2:48am UTC](https://discuss.pynq.io/t/fail-to-load-densebox/2200/1 "2021-01-04T02:48:04Z")

</div>

My environment:  
Board: ZCU104  
Image: pynq 2.6  
vitis-ai: docker 1.2.82  
pynq-dpu 1.2

Description:  
I can compile [cf\_densebox\_wider\_320\_320\_0.49G\_1.2](https://github.com/Xilinx/Vitis-AI/tree/v1.2.1/AI-Model-Zoo) successfully, and get the information as shown below.

 ![densebox_compile](https://us1.discourse-cdn.com/flex019/uploads/pynq1/original/2X/e/e3ac0f23ce935617d083ba554c385dd7710e687b.png)

When I load the model from the previous step and feed into dnndk by

```nohighlight
from dnndk import n2cube
overlay = DpuOverlay("dpu.bit")
overlay.load_model("./DL_model/dpu_densebox.elf")
n2cube.dpuOpen()
kernel = n2cube.dpuLoadKernel("densebox")

```

When I run it through jupyter notebook, I meet the problem like this.

 ![kernel_restarting](https://us1.discourse-cdn.com/flex019/uploads/pynq1/original/2X/0/0e52308b80922a7ad5ef7cd5dfccb3bfff4d791c.png)

BTW, I can run the dpu\_yolo\_v3 example in pynq-dpu normally.

---

<div class="post-metadata">

**Author:** ![pkchen](https://sea1.discourse-cdn.com/flex019/user_avatar/discuss.pynq.io/pkchen/32/1635_2.png) [@pkchen](https://discuss.pynq.io/u/pkchen)\
**Post date:** [January 11, 2021, 5:47am UTC](https://discuss.pynq.io/t/fail-to-load-densebox/2200/2 "2021-01-11T05:47:57Z")

</div>

Supplementary description:  
My problem is that I set up the environment according to official documents, and also compile the model successfully. Is there anything I need to modify to load the model through dnndk?

Since I can load and infer the same model through VART, the model should be fine.