search

hujiecpp / ISTR

ISTR: End-to-End Instance Segmentation with Transformers (https://arxiv.org/abs/2105.00637)
203 stars 28 forks source link

pred_mask visualization cannot match the object #4

Closed GuQingtao closed 3 years ago

GuQingtao commented 3 years ago

I trained the model on my dataset，when run the 'demo.py' for testing, I got the result as following: 1-01-5-pred It seems like that I got the right prediction result,but the code for visualization have some problems, or else??

luyao-cv commented 3 years ago

You used your own datasets. Have you re-extracted features from your datasets? Extract the new .npz file, and then train your data. In addition, I want to know how much box ap and mask ap of your training final model can reach?

GuQingtao commented 3 years ago

You used your own datasets. Have you re-extracted features from your datasets? Extract the new .npz file, and then train your data. In addition, I want to know how much box ap and mask ap of your training final model can reach?

I just run the train_net.py, and did not re-extracted features for my datasets to get the new .npz file , is it the reason leading to my result? Box: AP AP50 AP75 APs APm APl
96.741 99.787 98.564 nan nan 96.759

Mask: | 0.000 | 0.000 | 0.000 | nan | nan | 0.000 |

luyao-cv commented 3 years ago

I'm not sure if it is the only reason because in general, the performance is not so low, you can extract it first. In addition, you can use our trained weights to test and visualize your data.

GuQingtao commented 3 years ago

I have finished the problems,cause my own fault producing this error: I uesed detectron2 with pip install ,which cause a version error.Finally,use the detectron2 in this hub to fix it.

jingxinyi commented 2 years ago

I'm not sure if it is the only reason because in general, the performance is not so low, you can extract it first. In addition, you can use our trained weights to test and visualize your data.

Hello, I'm sorry that I can't find your trained weights. This error（AssertionError: Checkpoint ./output/model_final.pth not found!） occurred when I test.