Fine-tuning error

[Antonio Pardo de Santayana Navarro]

Hello, I am attempting to train tesseract using the tesstrain repo, but getting this error message as i make training.

make: * [Makefile:279: data/foo/checkpoints/foo_checkpoint] Violación de segmento (core dumped) [Núcleo vaciado a un archivo]

I also get some warnings along the way that might be indicative of the problem, such as:

Warning: LSTMTrainer deserialized an LSTMRecognizer!

Code range changed from 108 to 110!

Failed to read data from: ./data/foo/foo.wordlist

Failed to read data from: ./data/foo/foo.punc

Failed to read data from: ./data/foo/foo.numbers

Loaded unicharset of size 111 from file ./data/foo/unicharset

Bad box coordinates in boxfile string! DOÑANA 9

Extracting unicharset from plain text file ./data/foo/all-gt

Other case ñ of Ñ is not in unicharset

Other case u of U is not in unicharset

Other case v of V is not in unicharset

Other case k of K is not in unicharset

Other case f of F is not in unicharset

Other case b of B is not in unicharset

Other case W of w is not in unicharset

Other case j of J is not in unicharset

Other case p of P is not in unicharset

Other case x of X is not in unicharset

Other case z of Z is not in unicharset

Wrote unicharset file ./data/foo/my.unicharset

merge_unicharsets ./data/spa/foo.lstm-unicharset ./data/foo/my.unicharset  "data/foo/unicharset"

Loaded unicharset of size 109 from file ./data/spa/foo.lstm-unicharset

Loaded unicharset of size 56 from file ./data/foo/my.unicharset

My model is called foo (too lazy to change it), and im trying to tune from the spanish model. My makefile is as follows, excluding some uninteresting parts:

export

# Disable built-in suffix and implicit pattern rules (for software builds).

# This makes starting with a very large number of GT lines much faster.

MAKEFLAGS += -r

## Make sure that sort always uses the same sort order.

LC_ALL := C

SHELL := /bin/bash

LOCAL := $(PWD)/usr

PATH := $(LOCAL)/bin:$(PATH)

# Path to the .traineddata directory with traineddata suitable for training 

# (for example from tesseract-ocr/tessdata_best). Default: $(LOCAL)/share/tessdata

TESSDATA =  $(LOCAL)/share/tessdata

# Name of the model to be built. Default: $(MODEL_NAME)

MODEL_NAME = foo

# Data directory for output files, proto model, start model, etc. Default: $(DATA_DIR)

DATA_DIR = data

# Output directory for generated files. Default: $(OUTPUT_DIR)

OUTPUT_DIR = $(DATA_DIR)/$(MODEL_NAME)

# Ground truth directory. Default: $(GROUND_TRUTH_DIR)

GROUND_TRUTH_DIR := $(OUTPUT_DIR)-ground-truth

# Optional Wordlist file for Dictionary dawg. Default: $(WORDLIST_FILE)

WORDLIST_FILE := $(OUTPUT_DIR)/$(MODEL_NAME).wordlist

# Optional Numbers file for number patterns dawg. Default: $(NUMBERS_FILE)

NUMBERS_FILE := $(OUTPUT_DIR)/$(MODEL_NAME).numbers

# Optional Punc file for Punctuation dawg. Default: $(PUNC_FILE)

PUNC_FILE := $(OUTPUT_DIR)/$(MODEL_NAME).punc

# Name of the model to continue from. Default: '$(START_MODEL)'

START_MODEL = spa

LAST_CHECKPOINT = $(OUTPUT_DIR)/checkpoints/$(MODEL_NAME)_checkpoint

# Name of the proto model. Default: '$(PROTO_MODEL)'

PROTO_MODEL = $(OUTPUT_DIR)/$(MODEL_NAME).traineddata

# No of cores to use for compiling leptonica/tesseract. Default: $(CORES)

CORES = 4

# Leptonica version. Default: $(LEPTONICA_VERSION)

LEPTONICA_VERSION := 1.80.0

# Tesseract commit. Default: $(TESSERACT_VERSION)

TESSERACT_VERSION := 4.1.1

# Tesseract model repo to use. Default: $(TESSDATA_REPO)

TESSDATA_REPO = _best

# If EPOCHS is given, it is used to set MAX_ITERATIONS.

ifeq ($(EPOCHS),)

# Max iterations. Default: $(MAX_ITERATIONS)

MAX_ITERATIONS := 300

else

MAX_ITERATIONS := -$(EPOCHS)

endif

# Debug Interval. Default:  $(DEBUG_INTERVAL)

DEBUG_INTERVAL := 0

# Learning rate. Default: $(LEARNING_RATE)

ifdef START_MODEL

LEARNING_RATE := 0.0001

else

LEARNING_RATE := 0.002

endif

# Network specification. Default: $(NET_SPEC)

NET_SPEC := [1,36,0,1 Ct3,3,16 Mp3,3 Lfys48 Lfx96 Lrx96 Lfx192 O1c\#\#\#]

# Language Type - Indic, RTL or blank. Default: '$(LANG_TYPE)'

LANG_TYPE ?=

# Normalization mode - 2, 1 - for unicharset_extractor and Pass through Recoder for combine_lang_model

ifeq ($(LANG_TYPE),Indic)

NORM_MODE =2

RECODER =--pass_through_recoder

GENERATE_BOX_SCRIPT =generate_wordstr_box.py

else

ifeq ($(LANG_TYPE),RTL)

NORM_MODE =3

RECODER =--pass_through_recoder --lang_is_rtl

GENERATE_BOX_SCRIPT =generate_wordstr_box.py

else

NORM_MODE =2

RECODER=

GENERATE_BOX_SCRIPT =generate_line_box.py

endif

endif

# Page segmentation mode. Default: $(PSM)

PSM = 13

# Random seed for shuffling of the training data. Default: $(RANDOM_SEED)

RANDOM_SEED := 0

# Ratio of train / eval training data. Default: $(RATIO_TRAIN)

RATIO_TRAIN := 0.90

# Default Target Error Rate. Default: $(TARGET_ERROR_RATE)

TARGET_ERROR_RATE := 0.01

# BEGIN-EVAL makefile-parser --make-help Makefile

# END-EVAL

.PRECIOUS: $(OUTPUT_DIR)/checkpoints/$(MODEL_NAME)*_checkpoint

.PHONY: clean help leptonica lists proto-model tesseract tesseract-langs training unicharset

ALL_GT = $(OUTPUT_DIR)/all-gt

ALL_LSTMF = $(OUTPUT_DIR)/all-lstmf

# Create unicharset

unicharset: $(OUTPUT_DIR)/unicharset

# Create lists of lstmf filenames for training and eval

lists: $(OUTPUT_DIR)/list.train $(OUTPUT_DIR)/list.eval

$(OUTPUT_DIR)/list.eval \

$(OUTPUT_DIR)/list.train: $(ALL_LSTMF)

@mkdir -p $(OUTPUT_DIR)

@total=$$(wc -l < $(ALL_LSTMF)); \

  train=$$(echo "$$total * $(RATIO_TRAIN) / 1" | bc); \

  test "$$train" = "0" && \

    echo "Error: missing ground truth for training" && exit 1; \

  eval=$$(echo "$$total - $$train" | bc); \

  test "$$eval" = "0" && \

    echo "Error: missing ground truth for evaluation" && exit 1; \

  set -x; \

  head -n "$$train" $(ALL_LSTMF) > "$(OUTPUT_DIR)/list.train"; \

  tail -n "$$eval" $(ALL_LSTMF) > "$(OUTPUT_DIR)/list.eval"

ifdef START_MODEL

$(OUTPUT_DIR)/unicharset: $(ALL_GT)

@mkdir -p $(DATA_DIR)/$(START_MODEL)

combine_tessdata -u $(TESSDATA)/$(START_MODEL).traineddata  $(DATA_DIR)/$(START_MODEL)/$(MODEL_NAME)

unicharset_extractor --output_unicharset "$(OUTPUT_DIR)/my.unicharset" --norm_mode $(NORM_MODE) "$(ALL_GT)"

merge_unicharsets $(DATA_DIR)/$(START_MODEL)/$(MODEL_NAME).lstm-unicharset $(OUTPUT_DIR)/my.unicharset  "$@"

else

$(OUTPUT_DIR)/unicharset: $(ALL_GT)

@mkdir -p $(OUTPUT_DIR)

unicharset_extractor --output_unicharset "$@" --norm_mode $(NORM_MODE) "$(ALL_GT)"

endif

# Start training

training: $(OUTPUT_DIR).traineddata

$(ALL_GT): $(shell find -L $(GROUND_TRUTH_DIR) -name '*.gt.txt')

@mkdir -p $(OUTPUT_DIR)

find -L $(GROUND_TRUTH_DIR) -name '*.gt.txt' | xargs paste -s > "$@"

.PRECIOUS: %.box

%.box: %.png %.gt.txt

PYTHONIOENCODING=utf-8 python3 $(GENERATE_BOX_SCRIPT) -i "$*.png" -t "$*.gt.txt" > "$@"

%.box: %.bin.png %.gt.txt

PYTHONIOENCODING=utf-8 python3 $(GENERATE_BOX_SCRIPT) -i "$*.bin.png" -t "$*.gt.txt" > "$@"

%.box: %.nrm.png %.gt.txt

PYTHONIOENCODING=utf-8 python3 $(GENERATE_BOX_SCRIPT) -i "$*.nrm.png" -t "$*.gt.txt" > "$@"

%.box: %.tif %.gt.txt

PYTHONIOENCODING=utf-8 python3 $(GENERATE_BOX_SCRIPT) -i "$*.tif" -t "$*.gt.txt" > "$@"

$(ALL_LSTMF): $(patsubst %.gt.txt,%.lstmf,$(shell find -L $(GROUND_TRUTH_DIR) -name '*.gt.txt'))

@mkdir -p $(OUTPUT_DIR)

find -L $(GROUND_TRUTH_DIR) -name '*.lstmf' | python3 shuffle.py $(RANDOM_SEED) > "$@"

%.lstmf: %.box

@if test -f "$*.png"; then \

  image="$*.png"; \

elif test -f "$*.bin.png"; then \

  image="$*.bin.png"; \

elif test -f "$*.nrm.png"; then \

  image="$*.nrm.png"; \

else \

  image="$*.tif"; \

fi; \

set -x; \

tesseract "$${image}" $* --psm $(PSM) lstm.train

CHECKPOINT_FILES := $(wildcard $(OUTPUT_DIR)/checkpoints/$(MODEL_NAME)*.checkpoint)

.PHONY: traineddata

# Create best and fast .traineddata files from each .checkpoint file

traineddata: $(OUTPUT_DIR)/tessdata_best $(OUTPUT_DIR)/tessdata_fast

traineddata: $(subst checkpoints,tessdata_best,$(patsubst %.checkpoint,%.traineddata,$(CHECKPOINT_FILES)))

traineddata: $(subst checkpoints,tessdata_fast,$(patsubst %.checkpoint,%.traineddata,$(CHECKPOINT_FILES)))

$(OUTPUT_DIR)/tessdata_best $(OUTPUT_DIR)/tessdata_fast:

mkdir $@

$(OUTPUT_DIR)/tessdata_best/%.traineddata: $(OUTPUT_DIR)/checkpoints/%.checkpoint

lstmtraining \

          --stop_training \

          --continue_from $< \

          --traineddata $(PROTO_MODEL) \

          --model_output $@

$(OUTPUT_DIR)/tessdata_fast/%.traineddata: $(OUTPUT_DIR)/checkpoints/%.checkpoint

lstmtraining \

          --stop_training \

          --continue_from $< \

          --traineddata $(PROTO_MODEL) \

          --convert_to_int \

          --model_output $@

# Build the proto model

proto-model: $(PROTO_MODEL)

$(PROTO_MODEL): $(OUTPUT_DIR)/unicharset $(DATA_DIR)/radical-stroke.txt

combine_lang_model \

  --input_unicharset $(OUTPUT_DIR)/unicharset \

  --script_dir $(DATA_DIR) \

  --numbers $(NUMBERS_FILE) \

  --puncs $(PUNC_FILE) \

  --words $(WORDLIST_FILE) \

  --output_dir $(DATA_DIR) \

  $(RECODER) \

  --lang $(MODEL_NAME)

ifdef START_MODEL

$(LAST_CHECKPOINT): unicharset lists $(PROTO_MODEL)

@mkdir -p $(OUTPUT_DIR)/checkpoints

lstmtraining \

  --debug_interval $(DEBUG_INTERVAL) \

  --traineddata $(PROTO_MODEL) \

  --old_traineddata $(TESSDATA)/$(START_MODEL).traineddata \

  --continue_from $(DATA_DIR)/$(START_MODEL)/$(MODEL_NAME).lstm \

  --learning_rate $(LEARNING_RATE) \

  --model_output $(OUTPUT_DIR)/checkpoints/$(MODEL_NAME) \

  --train_listfile $(OUTPUT_DIR)/list.train \

  --eval_listfile $(OUTPUT_DIR)/list.eval \

  --max_iterations $(MAX_ITERATIONS) \

  --target_error_rate $(TARGET_ERROR_RATE)

$(OUTPUT_DIR).traineddata: $(LAST_CHECKPOINT)

lstmtraining \

--stop_training \

--continue_from $(LAST_CHECKPOINT) \

--traineddata $(PROTO_MODEL) \

--model_output $@

else

$(LAST_CHECKPOINT): unicharset lists $(PROTO_MODEL)

@mkdir -p $(OUTPUT_DIR)/checkpoints

lstmtraining \

  --debug_interval $(DEBUG_INTERVAL) \

  --traineddata $(PROTO_MODEL) \

  --learning_rate $(LEARNING_RATE) \

  --net_spec "$(subst c###,c`head -n1 $(OUTPUT_DIR)/unicharset`,$(NET_SPEC))" \

  --model_output $(OUTPUT_DIR)/checkpoints/$(MODEL_NAME) \

  --train_listfile $(OUTPUT_DIR)/list.train \

  --eval_listfile $(OUTPUT_DIR)/list.eval \

  --max_iterations $(MAX_ITERATIONS) \

  --target_error_rate $(TARGET_ERROR_RATE)

$(OUTPUT_DIR).traineddata: $(LAST_CHECKPOINT)

lstmtraining \

--stop_training \

--continue_from $(LAST_CHECKPOINT) \

--traineddata $(PROTO_MODEL) \

--model_output $@

endif

$(DATA_DIR)/radical-stroke.txt:

wget -O$@ 'https://github.com/tesseract-ocr/langdata_lstm/raw/master/radical-stroke.txt'

If anyone can help figure out where the error is coming from they'd save my life.