Skip to content

Commit da635bc

Browse files
yuwenzhojcwchen
yuwenzho
and
jcwchen
authored
add/update faster-rcnn, mask-rcnn, ssd-mobilenetv1, ssd and yolov3 qdq/int8 models (#587)
* add faster-rcnn, mask-rcnn, ssd-mobilenetv1 and ssd qdq models Signed-off-by: yuwenzho <[email protected]> * add/update faster-rcnn, mask-rcnn, ssd-mobilenetv1, ssd and yolov3 int8/qdq models Signed-off-by: yuwenzho <[email protected]> * reupload tar.gz file Signed-off-by: yuwenzho <[email protected]> --------- Signed-off-by: yuwenzho <[email protected]> Co-authored-by: Chun-Wei Chen <[email protected]>
1 parent 9e7f179 commit da635bc

24 files changed

+338
-61
lines changed

‎ONNX_HUB_MANIFEST.json‎

Lines changed: 272 additions & 27 deletions
Large diffs are not rendered by default.

‎vision/object_detection_segmentation/faster-rcnn/README.md‎

Lines changed: 5 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -11,8 +11,9 @@ This model is a real-time neural network for object detection that detects 80 di
1111
|-------------|:--------------|:--------------|:--------------|:--------------|:--------------|
1212
|Faster R-CNN R-50-FPN |[167.3 MB](model/FasterRCNN-10.onnx) |[158.0 MB](model/FasterRCNN-10.tar.gz) |1.5 |10 |mAP of 0.35 |
1313
|Faster R-CNN R-50-FPN-fp32 |[168.5 MB](model/FasterRCNN-12.onnx) |[156.2 MB](model/FasterRCNN-12.tar.gz) |1.9 |12 |mAP of 0.3437 |
14-
|Faster R-CNN R-50-FPN-int8 |[42.6 MB](model/FasterRCNN-12-int8.onnx) |[36.2 MB](model/FasterRCNN-12-int8.tar.gz) |1.9 |12 |mAP of 0.3399 |
15-
> Compared with the fp32 FasterRCNN-12, int8 FasterRCNN-12's mAP decline ratio is 1.11% and performance improvement is 1.43x.
14+
|Faster R-CNN R-50-FPN-int8 |[42.6 MB](model/FasterRCNN-12-int8.onnx) |[36.2 MB](model/FasterRCNN-12-int8.tar.gz) |1.9 |12 |mAP of 0.3409 |
15+
|Faster R-CNN R-50-FPN-qdq |[43 MB](model/FasterRCNN-12-qdq.onnx) |[29 MB](model/FasterRCNN-12-qdq.tar.gz) |1.9 |12 |mAP of 0.3390 |
16+
> Compared with the fp32 FasterRCNN-12, int8 FasterRCNN-12's mAP decline ratio is 0.81% and performance improvement is 1.43x.
1617
>
1718
> Note the performance depends on the test hardware.
1819
>
@@ -119,7 +120,7 @@ mAP of 0.353
119120
<hr>
120121

121122
## Quantization
122-
Faster R-CNN R-50-FPN-fp32 is obtained by quantizing Faster R-CNN R-50-FPN-fp32 model. We use [Intel® Neural Compressor](https://github.com/intel/neural-compressor) with onnxruntime backend to perform quantization. View the [instructions](https://github.com/intel/neural-compressor/blob/master/examples/onnxrt/object_detection/onnx_model_zoo/faster_rcnn/quantization/ptq/README.md) to understand how to use Intel® Neural Compressor for quantization.
123+
Faster R-CNN R-50-FPN-int8 and Faster R-CNN R-50-FPN-qdq are obtained by quantizing Faster R-CNN R-50-FPN-fp32 model. We use [Intel® Neural Compressor](https://github.com/intel/neural-compressor) with onnxruntime backend to perform quantization. View the [instructions](https://github.com/intel/neural-compressor/blob/master/examples/onnxrt/object_detection/onnx_model_zoo/faster_rcnn/quantization/ptq/README.md) to understand how to use Intel® Neural Compressor for quantization.
123124

124125
### Environment
125126
onnx: 1.9.0
@@ -153,6 +154,7 @@ Massa, Francisco and Girshick, Ross. maskrcnn-benchmark: Fast, modular reference
153154

154155
## Contributors
155156
* [mengniwang95](https://github.com/mengniwang95) (Intel)
157+
* [yuwenzho](https://github.com/yuwenzho) (Intel)
156158
* [airMeng](https://github.com/airMeng) (Intel)
157159
* [ftian1](https://github.com/ftian1) (Intel)
158160
* [hshen14](https://github.com/hshen14) (Intel)

‎vision/object_detection_segmentation/faster-rcnn/model/FasterRCNN-12-int8.onnx‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -1,3 +1,3 @@
11
version https://git-lfs.github.com/spec/v1
2-
oid sha256:c6bb169f20bd4bf1c08212f2c8b2693a22a6343454f8115fd8c0fadd2a49922a
3-
size 44626453
2+
oid sha256:95f67f5f6249f4804f1302367dd88cee32bf47713b9858cc6d8ba835548f9b8e
3+
size 44631113

‎vision/object_detection_segmentation/faster-rcnn/model/FasterRCNN-12-int8.tar.gz‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -1,3 +1,3 @@
11
version https://git-lfs.github.com/spec/v1
2-
oid sha256:d497a293e8486ec0a56f5276eb246b93280b7f31dc4e23a33db444d262384591
3-
size 38019008
2+
oid sha256:9edf754e3e4e6e78aed9a9d5690299edeba27b17d4dd43635a4fbd4c0e747be0
3+
size 38497565

‎vision/object_detection_segmentation/faster-rcnn/model/FasterRCNN-12-qdq.onnx‎

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,3 @@
1+
version https://git-lfs.github.com/spec/v1
2+
oid sha256:7b8b530b50b2fa9737625f2bbf454c439e9156c94d44e22602946d019f105f1c
3+
size 44525748

‎vision/object_detection_segmentation/faster-rcnn/model/FasterRCNN-12-qdq.tar.gz‎

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,3 @@
1+
version https://git-lfs.github.com/spec/v1
2+
oid sha256:90673a5a499343863af4c3f4efbed7e35c7a6c587dc37f2d5d2bf0bc77b6865a
3+
size 29844855

‎vision/object_detection_segmentation/mask-rcnn/README.md‎

Lines changed: 5 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -11,8 +11,9 @@ This model is a real-time neural network for object instance segmentation that d
1111
|-------------|:--------------|:--------------|:--------------|:--------------|:--------------|
1212
|Mask R-CNN R-50-FPN |[177.9 MB](model/MaskRCNN-10.onnx) | [168.8 MB](model/MaskRCNN-10.tar.gz) |1.5 |10 |mAP of 0.36 & 0.33 |
1313
|Mask R-CNN R-50-FPN-fp32 |[169.7 MB](model/MaskRCNN-12.onnx) | [157.3 MB](model/MaskRCNN-12.tar.gz) |1.9 |12 |mAP of 0.3372 |
14-
|Mask R-CNN R-50-FPN-int8 |[45.9 MB](model/MaskRCNN-12-int8.onnx) | [34.2 MB](model/MaskRCNN-12-int8.tar.gz) |1.9 |12 |mAP of 0.3340 |
15-
> Compared with the Mask R-CNN R-50-FPN-fp32, Mask R-CNN R-50-FPN-int8's mAP decline ratio is 0.95% and performance improvement is 1.99x.
14+
|Mask R-CNN R-50-FPN-int8 |[44 MB](model/MaskRCNN-12-int8.onnx) | [38 MB](model/MaskRCNN-12-int8.tar.gz) |1.9 |12 |mAP of 0.3314 |
15+
|Mask R-CNN R-50-FPN-qdq |[44 MB](model/MaskRCNN-12-qdq.onnx) | [30 MB](model/MaskRCNN-12-qdq.tar.gz) |1.9 |12 |mAP of 0.3328 |
16+
> Compared with the Mask R-CNN R-50-FPN-fp32, Mask R-CNN R-50-FPN-int8's mAP decline is 0.0058 and performance improvement is 1.99x.
1617
>
1718
> Note the performance depends on the test hardware.
1819
>
@@ -158,7 +159,7 @@ Metric is COCO box mAP@[IoU=0.50:0.95 | area=all | maxDets=100], computed over 2
158159
<hr>
159160

160161
## Quantization
161-
Mask R-CNN R-50-FPN-int8 is obtained by quantizing Mask R-CNN R-50-FPN-fp32 model. We use [Intel® Neural Compressor](https://github.com/intel/neural-compressor) with onnxruntime backend to perform quantization. View the [instructions](https://github.com/intel/neural-compressor/blob/master/examples/onnxrt/object_detection/onnx_model_zoo/mask_rcnn/quantization/ptq/README.md) to understand how to use Intel® Neural Compressor for quantization.
162+
Mask R-CNN R-50-FPN-int8 and Mask R-CNN R-50-FPN-qdq are obtained by quantizing Mask R-CNN R-50-FPN-fp32 model. We use [Intel® Neural Compressor](https://github.com/intel/neural-compressor) with onnxruntime backend to perform quantization. View the [instructions](https://github.com/intel/neural-compressor/blob/master/examples/onnxrt/object_detection/onnx_model_zoo/mask_rcnn/quantization/ptq/README.md) to understand how to use Intel® Neural Compressor for quantization.
162163

163164
### Environment
164165
onnx: 1.9.0
@@ -192,6 +193,7 @@ Massa, Francisco and Girshick, Ross. maskrcnn-benchmark: Fast, modular reference
192193

193194
## Contributors
194195
* [mengniwang95](https://github.com/mengniwang95) (Intel)
196+
* [yuwenzho](https://github.com/yuwenzho) (Intel)
195197
* [airMeng](https://github.com/airMeng) (Intel)
196198
* [ftian1](https://github.com/ftian1) (Intel)
197199
* [hshen14](https://github.com/hshen14) (Intel)

‎vision/object_detection_segmentation/mask-rcnn/model/MaskRCNN-12-int8.onnx‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -1,3 +1,3 @@
11
version https://git-lfs.github.com/spec/v1
2-
oid sha256:96388cfb596cbdb9a38dca16f1c22a1533843a030b094cdea3f1a2c90ef79156
3-
size 48137510
2+
oid sha256:4409935e855719fd6cd986f7ec2a3de840d0bd9c9cf7a0cba84ce95377f5b476
3+
size 45769352

‎vision/object_detection_segmentation/mask-rcnn/model/MaskRCNN-12-int8.tar.gz‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -1,3 +1,3 @@
11
version https://git-lfs.github.com/spec/v1
2-
oid sha256:edd066f26976b565171ea60563b6ec21a46f5ef5dc64592a4a33503089812918
3-
size 35901397
2+
oid sha256:0b16a7859d04601a3633a22c07797c9975b877ff384af23d7d9698467d57d333
3+
size 39438079

‎vision/object_detection_segmentation/mask-rcnn/model/MaskRCNN-12-qdq.onnx‎

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,3 @@
1+
version https://git-lfs.github.com/spec/v1
2+
oid sha256:d6fe3a9f49c3009d745b388c09c06ec3b332d942ba456053411a1f1bff22af9c
3+
size 45676737

0 commit comments

Comments
 (0)