Comments (11)
If the parameter cls_threshold_to_reg
is greater than zero, pretrain model should be used. If pretrain is not added, set this parameter to -1, which has little effect on the effect.
from sparse4d.
Yes, I have utilized the pre-trained model "fcos3d.pth" but I still meet the error. Can you train "sparse4d_r101_H1.py" without meeting any error?
from sparse4d.
I wrote some code to prevent alerting the bug and print a warning text every time it meets this case. And find that even in the 24th epoch I can still meet this case.
from sparse4d.
I did not encounter this problem. Is your batch_size not large enough? I used 8gpus*1sample_per_gpu.
from sparse4d.
It shouldn’t be considered a bug if all of them are false. Some samples may not have valid object, as long as it does not affect the convergence of loss
from sparse4d.
If you have concerns, just set this parameter to -1. It will not have a significant impact on the model effect
from sparse4d.
Yes, you are right. The reason I open this issue is that when I am evaluating the model at the 24th epoch, the mAOE is weird, as shown below. So I'm just wondering if this case causes the weird result.
mAP: 0.3939
mATE: 0.6969
mASE: 0.7060
mAOE: 1.5512
mAVE: 0.7891
mAAE: 0.2022
NDS: 0.3575
Eval time: 119.0s
Per-class results:
Object Class AP ATE ASE AOE AVE AAE
car 0.590 0.488 0.748 1.590 0.924 0.205
truck 0.323 0.744 0.791 1.588 0.837 0.217
bus 0.407 0.792 0.851 1.551 1.694 0.358
trailer 0.174 1.065 0.834 1.617 0.518 0.090
construction_vehicle 0.085 1.105 0.698 1.537 0.180 0.389
pedestrian 0.471 0.632 0.336 1.574 0.448 0.189
motorcycle 0.394 0.655 0.790 1.548 1.232 0.156
bicycle 0.306 0.620 0.807 1.704 0.480 0.015
traffic_cone 0.611 0.414 0.316 nan nan nan
barrier 0.577 0.454 0.888 1.251 nan nan
from sparse4d.
This phenomenon seems to be a problem with the annotation of yaw. It is recommended that you visualize the GT of the training data.
from sparse4d.
OK, thank you for your timely reply. I will close this issue.
from sparse4d.
hi,I meet the same
Yes, you are right. The reason I open this issue is that when I am evaluating the model at the 24th epoch, the mAOE is weird, as shown below. So I'm just wondering if this case causes the weird result.
mAP: 0.3939 mATE: 0.6969 mASE: 0.7060 mAOE: 1.5512 mAVE: 0.7891 mAAE: 0.2022 NDS: 0.3575 Eval time: 119.0s Per-class results: Object Class AP ATE ASE AOE AVE AAE car 0.590 0.488 0.748 1.590 0.924 0.205 truck 0.323 0.744 0.791 1.588 0.837 0.217 bus 0.407 0.792 0.851 1.551 1.694 0.358 trailer 0.174 1.065 0.834 1.617 0.518 0.090 construction_vehicle 0.085 1.105 0.698 1.537 0.180 0.389 pedestrian 0.471 0.632 0.336 1.574 0.448 0.189 motorcycle 0.394 0.655 0.790 1.548 1.232 0.156 bicycle 0.306 0.620 0.807 1.704 0.480 0.015 traffic_cone 0.611 0.414 0.316 nan nan nan barrier 0.577 0.454 0.888 1.251 nan nan
hi,I met the same problem at the 24th epoch,have you solved it?
from sparse4d.
If you are using the code of sparse4dv1 or sparse4dv2, update the version of mmdet3d >= rc1.0.0. @bobyang7
from sparse4d.
Related Issues (20)
- Tracking Pipeline of Sparse4Dv3 HOT 1
- 请问可以提供checkpoint吗 HOT 1
- 关于深度 HOT 3
- Inquiries about ckpt files learned by VoVNet and ResNet-101 backbone on the Sparse4Dv2 HOT 3
- When will sparse4d v3 be released ? HOT 3
- About mAOE in r50 HOT 2
- 【关于DenseDepthNet】如何设计密集深度预测的监督方式? HOT 2
- 您好,我在复现您的工作的时候,出现了NuScenes3DDetTrackDataset Not Found的问题 HOT 3
- Suggested training configurations for smaller batch sizes? HOT 2
- Why is learning rate in R50 logs half of the config lr? HOT 2
- Regarding Engineering Applications HOT 1
- Inference issue HOT 1
- 你们公司招聘远程工作者吗,刷nuscenes数据集的 HOT 1
- NMS implementation for multiple false positive HOT 30
- raise Exception('Error: Invalid box type: %s' % box)
- 关于速度预测逻辑 HOT 2
- 请教一个关于v3的tracking逻辑问题 HOT 1
- How was this file generated? nuscenes_infos_test.pkl include nuscenes_infos_train、nuscenes_infos_test
- Pretrain Weights mismatch?
- How did you get nuscenes_cam and the pkl files inside?
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from sparse4d.