singa-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "zhangzhaoqi (Jira)" <j...@apache.org>
Subject [jira] [Updated] (SINGA-509) Add opertors for all onnx examples
Date Mon, 04 May 2020 07:25:00 GMT

     [ https://issues.apache.org/jira/browse/SINGA-509?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

zhangzhaoqi updated SINGA-509:
------------------------------
    Description: 
As we have supported some ONNX examples, we hope to support all ONNX examples.

 

So I summary the operators which we need to support as following:

 
 * c++ means: the implement in python will be inefficient, may need to move to c++ to implement.
 * invalid means: cannot find in ONNX operators set.
 * hard means: hard to implement, need to discuss further.

 
|total|workload|bidaf|gpt2|fast_neural_style|super_resolution|ResNet101_DUC_HDC|yolov3-tiny|yolov3|yolov2_coco|yolov2|mask_rcnn_R_50_FPN_1x|faster_rcnn_R_50_FPN_1x|ssd|zfnet|ShuffleNet_V2|ShuffleNet_V1|inception_v2|inception_v1|densenet121|bvlc_reference_rcnn_ilsvrc13|bvlc_reference_caffenet|bvlc_googlenet|bvlc_alexnet|squeezenet|
|Abs|1h|√| | | | | | | | | | | | | | | | | | | | | | |
|ArgMax|c++|√| | | | | | | | | | | | | | | | | | | | | | |
|CategoryMapper|invalid|√| | | | | | | | | | | | | | | | | | | | | | |
|Compress|c++|√| | | | | | | | | | | | | | | | | | | | | | |
|ConvTranspose|6h| | | | | | | | | |√| | | | | | | | | | | | | |
|Equal|1h| | | | | | | | | |√|√| | | | | | | | | | | | |
|Exp|1h| | | | | |√|√| | |√|√|√| | | | | | | | | | | |
|Expand|6h| | | | | | | | | |√|√| | | | | | | | | | | | |
|Floor|1h| | |√| | | | | | |√|√| | | | | | | | | | | | |
|Hardmax|3h|√| | | | | | | | | | | | | | | | | | | | | | |
|ImageScaler|invalid| | | | | | | | |√| | | | | | | | | | | | | | |
|InstanceNormalization|c++| | |√| | | | | | | | | | | | | | | | | | | | |
|LRN|hard| | | | | | | | | | | | |√| | | |√| |√|√|√|√| |
|LSTM|12h|√| | | | | | | | | | | | | | | | | | | | | | |
|Loop|12h| | | | | | | | | | | | | | | | | | | | | | | |
|NonMaxSuppression|12h| | | | | |√|√| | |√|√|√| | | | | | | | | | | |
|Pad|3h| | |√| | | | | | | | | | | | | | | | | | | | |
|ReduceMax|c++|√| | | | | | | | | | | | | | | | | | | | | | |
|ReduceMin|c++| | | | | |√|√| | |√|√|√| | | | | | | | | | | |
|Resize|12h| | | | | |√|√| | |√|√| | | | | | | | | | | | |
|RoiAlign|6h| | | | | | | | | |√|√| | | | | | | | | | | | |
|Round|1h| | | | | |√| | | | | | | | | | | | | | | | | |
|Scan|hard|√| | | | | | | | | | | | | | | | | | | | | | |
|ScatterElements|c++| | | | | | | | | |√|√| | | | | | | | | | | | |
|SpaceToDepth|c++| | | | | | | | |√| | | | | | | | | | | | | | |
|TopK|c++| | | | | | | | | |√|√|√| | | | | | | | | | | |
|Upsample|6h| | |√| | | | | | | | | | | | | | | | | | | | |

  was:
As we have supported some ONNX examples, we hope to support all ONNX examples.

 

So I summary the operators which we need to support as following:

 
 * c++ means: the implement in python will be inefficient, may need to move to c++ to implement,
 * 

 
|total|workload|bidaf|gpt2|fast_neural_style|super_resolution|ResNet101_DUC_HDC|yolov3-tiny|yolov3|yolov2_coco|yolov2|mask_rcnn_R_50_FPN_1x|faster_rcnn_R_50_FPN_1x|ssd|zfnet|ShuffleNet_V2|ShuffleNet_V1|inception_v2|inception_v1|densenet121|bvlc_reference_rcnn_ilsvrc13|bvlc_reference_caffenet|bvlc_googlenet|bvlc_alexnet|squeezenet|
|Abs|1h|√| | | | | | | | | | | | | | | | | | | | | | |
|ArgMax|c++|√| | | | | | | | | | | | | | | | | | | | | | |
|CategoryMapper|invalid|√| | | | | | | | | | | | | | | | | | | | | | |
|Compress|c++|√| | | | | | | | | | | | | | | | | | | | | | |
|ConvTranspose|6h| | | | | | | | | |√| | | | | | | | | | | | | |
|Equal|1h| | | | | | | | | |√|√| | | | | | | | | | | | |
|Exp|1h| | | | | |√|√| | |√|√|√| | | | | | | | | | | |
|Expand|6h| | | | | | | | | |√|√| | | | | | | | | | | | |
|Floor|1h| | |√| | | | | | |√|√| | | | | | | | | | | | |
|Hardmax|3h|√| | | | | | | | | | | | | | | | | | | | | | |
|ImageScaler|invalid| | | | | | | | |√| | | | | | | | | | | | | | |
|InstanceNormalization|c++| | |√| | | | | | | | | | | | | | | | | | | | |
|LRN|hard| | | | | | | | | | | | |√| | | |√| |√|√|√|√| |
|LSTM|12h|√| | | | | | | | | | | | | | | | | | | | | | |
|Loop|12h| | | | | | | | | | | | | | | | | | | | | | | |
|NonMaxSuppression|12h| | | | | |√|√| | |√|√|√| | | | | | | | | | | |
|Pad|3h| | |√| | | | | | | | | | | | | | | | | | | | |
|ReduceMax|c++|√| | | | | | | | | | | | | | | | | | | | | | |
|ReduceMin|c++| | | | | |√|√| | |√|√|√| | | | | | | | | | | |
|Resize|12h| | | | | |√|√| | |√|√| | | | | | | | | | | | |
|RoiAlign|6h| | | | | | | | | |√|√| | | | | | | | | | | | |
|Round|1h| | | | | |√| | | | | | | | | | | | | | | | | |
|Scan|hard|√| | | | | | | | | | | | | | | | | | | | | | |
|ScatterElements|c++| | | | | | | | | |√|√| | | | | | | | | | | | |
|SpaceToDepth|c++| | | | | | | | |√| | | | | | | | | | | | | | |
|TopK|c++| | | | | | | | | |√|√|√| | | | | | | | | | | |
|Upsample|6h| | |√| | | | | | | | | | | | | | | | | | | | |


> Add opertors for all onnx examples
> ----------------------------------
>
>                 Key: SINGA-509
>                 URL: https://issues.apache.org/jira/browse/SINGA-509
>             Project: Singa
>          Issue Type: New Feature
>            Reporter: zhangzhaoqi
>            Priority: Major
>
> As we have supported some ONNX examples, we hope to support all ONNX examples.
>  
> So I summary the operators which we need to support as following:
>  
>  * c++ means: the implement in python will be inefficient, may need to move to c++ to
implement.
>  * invalid means: cannot find in ONNX operators set.
>  * hard means: hard to implement, need to discuss further.
>  
> |total|workload|bidaf|gpt2|fast_neural_style|super_resolution|ResNet101_DUC_HDC|yolov3-tiny|yolov3|yolov2_coco|yolov2|mask_rcnn_R_50_FPN_1x|faster_rcnn_R_50_FPN_1x|ssd|zfnet|ShuffleNet_V2|ShuffleNet_V1|inception_v2|inception_v1|densenet121|bvlc_reference_rcnn_ilsvrc13|bvlc_reference_caffenet|bvlc_googlenet|bvlc_alexnet|squeezenet|
> |Abs|1h|√| | | | | | | | | | | | | | | | | | | | | | |
> |ArgMax|c++|√| | | | | | | | | | | | | | | | | | | | | | |
> |CategoryMapper|invalid|√| | | | | | | | | | | | | | | | | | | | | | |
> |Compress|c++|√| | | | | | | | | | | | | | | | | | | | | | |
> |ConvTranspose|6h| | | | | | | | | |√| | | | | | | | | | | | | |
> |Equal|1h| | | | | | | | | |√|√| | | | | | | | | | | | |
> |Exp|1h| | | | | |√|√| | |√|√|√| | | | | | | | | | | |
> |Expand|6h| | | | | | | | | |√|√| | | | | | | | | | | | |
> |Floor|1h| | |√| | | | | | |√|√| | | | | | | | | | | | |
> |Hardmax|3h|√| | | | | | | | | | | | | | | | | | | | | | |
> |ImageScaler|invalid| | | | | | | | |√| | | | | | | | | | | | | | |
> |InstanceNormalization|c++| | |√| | | | | | | | | | | | | | | | | | | | |
> |LRN|hard| | | | | | | | | | | | |√| | | |√| |√|√|√|√| |
> |LSTM|12h|√| | | | | | | | | | | | | | | | | | | | | | |
> |Loop|12h| | | | | | | | | | | | | | | | | | | | | | | |
> |NonMaxSuppression|12h| | | | | |√|√| | |√|√|√| | | | | | | | | | | |
> |Pad|3h| | |√| | | | | | | | | | | | | | | | | | | | |
> |ReduceMax|c++|√| | | | | | | | | | | | | | | | | | | | | | |
> |ReduceMin|c++| | | | | |√|√| | |√|√|√| | | | | | | | | | | |
> |Resize|12h| | | | | |√|√| | |√|√| | | | | | | | | | | | |
> |RoiAlign|6h| | | | | | | | | |√|√| | | | | | | | | | | | |
> |Round|1h| | | | | |√| | | | | | | | | | | | | | | | | |
> |Scan|hard|√| | | | | | | | | | | | | | | | | | | | | | |
> |ScatterElements|c++| | | | | | | | | |√|√| | | | | | | | | | | | |
> |SpaceToDepth|c++| | | | | | | | |√| | | | | | | | | | | | | | |
> |TopK|c++| | | | | | | | | |√|√|√| | | | | | | | | | | |
> |Upsample|6h| | |√| | | | | | | | | | | | | | | | | | | | |



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

Mime
View raw message