search

Belval / CRNN

A TensorFlow implementation of https://github.com/bgshih/crnn
MIT License
297 stars 101 forks source link

Default Model gives poor testing result. #65

Open shivamkumarsingh114 opened 4 years ago

shivamkumarsingh114 commented 4 years ago

I am using the dataset specified by you with images in the format label_index.jpg. However, testing results are very poor. Dataset being used is: https://node1.belval.org/test_dataset.zip. I stored it in data folder and extracted the images.

Command being executed: python3 run.py -ex ./data/ --test --restore -bs 32

Output: Use standard file APIs to check for files with this prefix. Loading data Testing beignet iiiii apocynaceous iiiii arshine iiiii

Why is the output always "iiiiii"?

Timthony commented 4 years ago

I am using the dataset specified by you with images in the format label_index.jpg. However, testing results are very poor. Dataset being used is: https://node1.belval.org/test_dataset.zip. I stored it in data folder and extracted the images.

Command being executed: python3 run.py -ex ./data/ --test --restore -bs 32

Output: Use standard file APIs to check for files with this prefix. Loading data Testing beignet iiiii apocynaceous iiiii arshine iiiii

Why is the output always "iiiiii"?

I have the same problem. Have you solved it?

Timthony commented 4 years ago

I have the same problem. Have you solved it?

shivamkumarsingh114 commented 4 years ago

I trained the model from scratch. Seems to be working but still not as satisfactory as CRNN should be. I am thinking of rewriting the library from scratch.

Belval commented 4 years ago

Indeed there seems to be a problem somewhere as I cannot train the model or get suitable predictions anymore. I'll investigate but since the code is in TF 1 I might just rewrite it in eager mode.

shivamkumarsingh114 commented 4 years ago

Did you identify the problem?