You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardexpand all lines: README.md
+44-2
Original file line number
Diff line number
Diff line change
@@ -1,4 +1,4 @@
1
-
# YOLOv9 QAT for TensorRT
1
+
# YOLOv9 QAT for TensorRT Detection / Segmentation
2
2
3
3
This repository contains an implementation of YOLOv9 with Quantization-Aware Training (QAT), specifically designed for deployment on platforms utilizing TensorRT for hardware-accelerated inference. <br>
4
4
This implementation aims to provide an efficient, low-latency version of YOLOv9 for real-time detection applications.<br>
@@ -15,7 +15,8 @@ We use [TensorRT's pytorch quntization tool](https://github.com/NVIDIA/TensorRT/
15
15
For those who are not familiar with QAT, I highly recommend watching this video:<br> [Quantization explained with PyTorch - Post-Training Quantization, Quantization-Aware Training](https://www.youtube.com/watch?v=0VdNflU08yA)
16
16
17
17
**Important**<br>
18
-
Currently, quantization is only available for object detection models. However, since quantization primarily affects the backbone of the YOLOv9 model and the backbone remains consistent across all YOLOv9 variants, quantization is effectively prepared for all YOLOv9-based models, regardless of whether they are used for detection or segmentation tasks. Quantization support for segmentation models has not yet been released, as it necessitates the development of evaluation criteria and the validation of quantization for the final layers of the model. <br>
18
+
Evaluation of the segmentation model using TensorRT is currently under development. Once I have more available time, I will complete and release this work.
19
+
19
20
🌟 We still have plenty of nodes to improve Q/DQ, and we rely on the community's contribution to enhance this project, benefiting us all. Let's collaborate and make it even better! 🚀
20
21
21
22
## Release Highlights
@@ -35,6 +36,7 @@ Currently, quantization is only available for object detection models. However,
35
36
36
37
### Evaluation Results
37
38
39
+
## Detection
38
40
#### Activation SiLU
39
41
40
42
| Eval Model | AP | AP50 | Precision | Recall |
@@ -66,6 +68,14 @@ Currently, quantization is only available for object detection models. However,
66
68
|**INT8 (TensorRT)** vs **Origin (Pytorch)**|||||
67
69
|| -0.002 | -0.005 | +0.004 | -0.003 |
68
70
71
+
## Segmentation
72
+
| Model | Box |||| Mask ||||
73
+
|--------|-----|--|--|--|------|--|--|--|
74
+
|| P | R | mAP50 | mAP50-95 | P | R | mAP50 | mAP50-95 |
parser.add_argument('--mask-resolution', type=int, default=160, help='Mask pooled output.')
773
+
parser.add_argument('--pooler-scale', type=float, default=0.25, help='Multiplicative factor used to translate the ROI coordinates. ')
774
+
parser.add_argument('--sampling-ratio', type=int, default=0, help='Number of sampling points in the interpolation. Allowed values are non-negative integers.')
0 commit comments