training issues IOU_K always on 0.5000 and loss around 1.000 - Githubissues

whai362 / PSENet

Official Pytorch implementations of PSENet.

Apache License 2.0

1.17k stars 344 forks source link

training issues IOU_K always on 0.5000 and loss around 1.000 #168

Closed kbrajwani closed 3 years ago

kbrajwani commented 3 years ago

i am doing training on colab with python 3.6 with latest pytorch version. dataset is icdar 15 train_ic15.py script that i am running

(1/125) Batch: 8.914s | TOTAL: 0min | ETA: 19min | Loss: 1.0000 | Acc_t: 0.1929 | IOU_t: 0.1929 | IOU_k: 0.5000 (21/125) Batch: 2.421s | TOTAL: 1min | ETA: 4min | Loss: 1.0000 | Acc_t: 0.1954 | IOU_t: 0.1954 | IOU_k: 0.5000 (41/125) Batch: 2.354s | TOTAL: 2min | ETA: 3min | Loss: 1.0000 | Acc_t: 0.1944 | IOU_t: 0.1944 | IOU_k: 0.5000

can you tell me why this is happening?