Deserialized Header Failed Mac OS X

[MichaelLLi]

Environment

• Tesseract Version: 4.1.1
• Commit Number: 7510304
• Platform: Mac OS X 10.14

Current Behavior:

So I installed 4.1.1 along with the training tools from source and decided to try out the whole training process. I downloaded a sample TIFF file from the internet (CCITT_1.tiff) and then created the box/tiff file pair as:

tesseract CCITT_1.TIFF CCITT lstmbox

This gave:

Tesseract Open Source OCR Engine v4.1.1 with Leptonica
Page 1

With no error.

Then I created the lstmf files as:

tesseract CCITT_1.TIFF lstmf/valid1 lstm.train

And it gave me the same as above with no error. Finally, I ran a dummy training script with just this one picture:

lstmtraining --model_output models/test --continue_from models/eng.lstm --traineddata data/eng-best.traineddata --train_listfile dataset/annotated/training_files.txt --max_iterations 100

And I get:

Loaded file models/eng.lstm, unpacking...
Warning: LSTMTrainer deserialized an LSTMRecognizer!
Continuing from models/eng.lstm
Deserialize header failed: /Users/Mlli/Documents/GitHub/LineageEyeMachineLearning/ocr/lstmf/valid1.lstmf
Load of page 0 failed!
Load of images failed!!

I have made sure that the files do actually exist at the location, and the eng-best.traineddata is the LSTM only one. What could be wrong here? Thanks!

Expected Behavior:

Suggested Fix:

[MichaelLLi]

Tesseract.zip

The relevant files are attached here.

[Shreeshrii]

Is the file there at the path specified. What's the output of

ls -l /Users/Mlli/Documents/GitHub/LineageEyeMachineLearning/ocr/lstmf/valid1.lstmf

In the zipfile that you have attached, the lstmf file is in the main directory (not in subdirectory called lstmf).

[amitdo]

No response from the OP.