NPU documentation and tools

VIM3
21

Same here. Waiting for the NPU doc and examples before buying boards.

1 Like
22

Processing.

Sorry for the late response.

23

We will finally release the toolkit in public, but it just take a few more days for that :slight_smile:

For the users who want to get the the NPU toolkit and documents earlier, have to apply by email first.

Good day!

24

Hi Gouwa,

which email?

1 Like
25

I bought two vim 3 pro and also I’m interested in the NPU toolkit/documentation, both for bench-marking and implementation in my projects. I’ll follow up with an email.

1 Like
26

Also do you have any plan to support Tensorflow or Pytorch with NPU/GPU? That’s also one of the reason why I invest on this sbc.

2 Likes
27

Any updates on this topic??

28

Hi check here:

Thanks!

1 Like
29
1 Like
30

I have applied, no links so far
richard@optasia.com.sg
thanks

31

Already sent out, please check again.

32

Gouwa,

Verisilicon has released a lot of models in the github.
Can these models be used with NPU SDK? If yes, how?

Thanks,
Kaishi

33

@numbqq @Frank will confirm and response.

BTW, we will release a interesting demo on VIM3 soon :wink:

Good day!

34

@numbqq @Frank

Any comment on verisilicon acuity model?

Guawa,

When will new demo be available?

Thanks,

35

What model? You can convet any models you want.

In the end of this week we will release more demo.

1 Like
36

I’m getting an error while trying to convert mobilenet ssd v1 from Acuity Model zoo. Most of the classifiers seem to be convert ok, however SSD/object detectors seem to fail.

convertensorflow --tf-pb ./model/frozen_inference_graph.pb --inputs image_tensor --input-size-list ‘300 ,300,3’ --outputs ‘detection_classes detection_boxes detection_scores num_detections’ --net-output mobilenet_ssd_v1.json --data-output mobilenet_ssd_v1.data

Fails with error: …
D Try match Identity num_detections
I Match identity [[‘num_detections’]] [[‘Identity’]] to [[‘noop’]]
D Try match TensorArrayGatherV3 Postprocessor/BatchMultiClassNonMaxSuppression/map/TensorArrayStack/TensorArrayGatherV3
W Not match node Postprocessor/BatchMultiClassNonMaxSuppression/map/TensorArrayStack/TensorArrayGatherV3 TensorArrayGatherV3

Is acuity supporting mobilenet SSD ? I did a little bit of digging around, however I am not certain what is the root cause? could be Acuity lib/toolset dependency on tensorflow 1.10 (older version)? is there a new acuitylib available?

Thanks ,
Larry

37

@Frank Please follow up.

38

@larrylart Can you give me the pb file you used for ? And We need the error info . Maybe we can find something in it .

39

İ think thats cause Last layer “nonMaxSupressionLayer”. I Got error like this when i have converted tinyYolo3,.

40

Hi Frank, I used the .pb file from this archive: http://download.tensorflow.org/models/object_detection/ssd_mobilenet_v1_coco_2018_01_28.tar.gz

I uploaded the actual .pb file and the full console logs at https://github.com/larrylart/codrive/tree/master/hw_sampling/vim3_npu/debug_logs

I though this could have something to do with multiple subgraphs? I tried different configurations of --subgraphs but I am not even sure if that option does anything at all?

And the error:
D Try match Identity detection_classes
I Match identity [[‘detection_classes’]] [[‘Identity’]] to [[‘noop’]]
D Try match Identity detection_boxes
I Match identity [[‘detection_boxes’]] [[‘Identity’]] to [[‘noop’]]
D Try match Identity detection_scores
I Match identity [[‘detection_scores’]] [[‘Identity’]] to [[‘noop’]]
D Try match Identity num_detections
I Match identity [[‘num_detections’]] [[‘Identity’]] to [[‘noop’]]
D Try match Add add
I Match add [[‘add’]] [[‘Add’]] to [[‘add’]]
D Try match TensorArrayGatherV3 Postprocessor/BatchMultiClassNonMaxSuppression/map/TensorArrayStack/TensorArrayGatherV3
W Not match node Postprocessor/BatchMultiClassNonMaxSuppression/map/TensorArrayStack/TensorArrayGatherV3 TensorArrayGatherV3
Traceback (most recent call last):
File “convertensorflow.py”, line 62, in
File “convertensorflow.py”, line 58, in main
File “acuitylib/app/importer/import_tensorflow.py”, line 150, in run
File “acuitylib/converter/convert_tf.py”, line 584, in match_paragraph_and_param
File “acuitylib/converter/convert_tf.py”, line 483, in _tf_push_ready_node
TypeError: ‘NoneType’ object is not iterable
[59984] Failed to execute script convertensorflow