huawei-noah / Efficient-Computing

Efficient computing methods developed by Huawei Noah's Ark Lab
1.21k stars 211 forks source link

AssertionError: Results do not correspond to current coco set #86

Closed Egorundel closed 1 year ago

Egorundel commented 1 year ago

Hello, I have such a mistake when learning after the first epoch of learning. I teach on single GPU.

I run this command: python3 tools/train.py --batch 4 --conf configs/gold_yolo-l.py --data data/custom.yaml --epoch 3 --fuse_ab --device 0 --name gold_yolo-l Screenshot from 2023-10-17 10-23-08

Error: AssertionError: Results do not correspond to current coco set Screenshot from 2023-10-17 10-23-02

My config file of Gold-Yolo-L without SyncBN: Screenshot from 2023-10-17 10-41-47

lose4578 commented 1 year ago

Hi, whether you are using a custom data set. If yes, you should set the is_coco: False in your data/custom.yaml

Egorundel commented 1 year ago

@lose4578 Hi! I have this data/custom.yaml: 2023_10_18_0m0_Kleki

I still have the same error

lose4578 commented 1 year ago

This looks like the code error in YOLOv6, maybe you could try to comment out anno_path, follow in this issuse https://github.com/meituan/YOLOv6/issues/48

Egorundel commented 1 year ago

@lose4578 Thank you, comrade! You are the best support!

Screenshot from 2023-10-18 15-41-56