Submission: Synthetica/YCB-V/Synthetica-ConvNeXt-S
Download submission
| Submission name |
Synthetica-ConvNeXt-S
|
| Submission time (UTC) |
March 4, 2024, 7:52 p.m.
|
| User |
quantumsprinter
|
| Task |
Model-based 2D detection of seen objects |
| Dataset |
YCB-V |
| Training image type |
Synthetic + real |
| Description |
We use ConvNeXt-S model as a backbone to RT-DETR. Omniverse Isaac Sim was used to render data.
TRT at fp32
|
| Evaluation scores |
| AP: | 0.885 |
| AP50: | 0.998 |
| AP75: | 0.988 |
| AP_large: | 0.918 |
| AP_medium: | 0.818 |
| AP_small: | 0.551 |
| AR1: | 0.903 |
| AR10: | 0.905 |
| AR100: | 0.905 |
| AR_large: | 0.937 |
| AR_medium: | 0.827 |
| AR_small: | 0.550 |
| average_time_per_image: | 0.020 |
|
| User |
quantumsprinter
|
| Publication |
|
| Implementation |
RT-DETR + OV synthetic data + real data + TensorRT |
| Training image modalities |
RGB |
| Test image modalities |
RGB |
| Description |
|
| Computer specifications |
3090 |