✅ Use GT segmentation to test

.circleci/config.yml

@@ -8,7 +8,7 @@ jobs:
     docker:
       - image: ubuntu:19.04
     steps:
-      - run: apt-get update ; apt-get install -y make git curl python3 python3-pip libtesseract-dev libleptonica-dev tesseract-ocr-eng tesseract-ocr wget
+      - run: apt-get update ; apt-get install -y make git curl python3 python3-pip wget
       - checkout
       - run: make install PIP_INSTALL="pip3 install"
       - run: pip3 install -r requirements-test.txt

Makefile

@@ -5,7 +5,7 @@ PYTHON = python
 PIP_INSTALL = pip install
 
 # '$(GIT_CLONE)'
-GIT_CLONE = git clone --depth 1
+GIT_CLONE = git clone
 
 # BEGIN-EVAL makefile-parser --make-help Makefile
 
@@ -42,7 +42,10 @@ calamari:
 
 # Clone calamari_models repo
 calamari_models:
-	$(GIT_CLONE) https://github.com/chwick/calamari_models
+	$(GIT_CLONE) -n https://github.com/chwick/calamari_models
+	# Checkout latest version that works with calamari-ocr==0.3.5:
+	git checkout f76b1d3ec
+
 
 # pip install calamari
 calamari/build: calamari calamari_models

requirements-test.txt

@@ -1,3 +1,2 @@
 pytest
-ocrd_tesserocr >= 0.4.0
 coverage

test/test_recognize.py

@@ -1,60 +1,50 @@
 import os
-from os.path import join, exists
 import shutil
+import urllib.request
 
 from test.base import TestCase, main, assets, skip
 
 from ocrd.resolver import Resolver
 
-from ocrd_tesserocr import TesserocrSegmentRegion
-from ocrd_tesserocr import TesserocrSegmentLine
-
 from ocrd_calamari import CalamariRecognize
 
-#METS_HEROLD_SMALL = assets.url_of('SBB0000F29300010000/data/mets_one_file.xml')
-# as long as #96 remains, we cannot use workspaces which have local relative files:
-METS_HEROLD_SMALL = assets.url_of('kant_aufklaerung_1784-binarized/data/mets.xml')
+METS_KANT = assets.url_of('kant_aufklaerung_1784-page-block-line-word_glyph/data/mets.xml')
 
 WORKSPACE_DIR = '/tmp/test-ocrd-calamari'
 
 class TestCalamariRecognize(TestCase):
 
     def setUp(self):
-        if exists(WORKSPACE_DIR):
+        if os.path.exists(WORKSPACE_DIR):
             shutil.rmtree(WORKSPACE_DIR)
         os.makedirs(WORKSPACE_DIR)
 
-    #skip("Takes too long")
     def runTest(self):
         resolver = Resolver()
-        workspace = resolver.workspace_from_url(METS_HEROLD_SMALL, dst_dir=WORKSPACE_DIR)
+        workspace = resolver.workspace_from_url(METS_KANT, dst_dir=WORKSPACE_DIR)
 
-        TesserocrSegmentRegion(
-            workspace,
-            input_file_grp="OCR-D-IMG",
-            output_file_grp="OCR-D-SEG-BLOCK"
-        ).process()
-        workspace.save_mets()
+        # XXX Work around data bug(?):
+        #     PAGE-XML links to OCR-D-IMG/INPUT_0017.tif, but this is nothing core can download
+        os.makedirs(os.path.join(WORKSPACE_DIR, 'OCR-D-IMG'))
+        for f in ['INPUT_0017.tif', 'INPUT_0020.tif']:
+            urllib.request.urlretrieve(
+                    "https://github.com/OCR-D/assets/raw/master/data/kant_aufklaerung_1784/data/OCR-D-IMG/" + f,
+                    os.path.join(WORKSPACE_DIR, 'OCR-D-IMG', f))
 
-        TesserocrSegmentLine(
-            workspace,
-            input_file_grp="OCR-D-SEG-BLOCK",
-            output_file_grp="OCR-D-SEG-LINE"
-        ).process()
-        workspace.save_mets()
+        # XXX Should remove GT text to really test this
 
         CalamariRecognize(
             workspace,
-            input_file_grp="OCR-D-SEG-LINE",
+            input_file_grp="OCR-D-GT-SEG-LINE",
             output_file_grp="OCR-D-OCR-CALAMARI",
             parameter={
-                'checkpoint': 'calamari_models/fraktur_historical/*.ckpt.json'
+                'checkpoint': os.path.join(os.getcwd(), 'calamari_models/fraktur_19th_century/*.ckpt.json')
             }
         ).process()
         workspace.save_mets()
 
-        page1 = join(workspace.directory, 'OCR-D-OCR-CALAMARI/OCR-D-OCR-CALAMARI_0001.xml')
-        self.assertTrue(exists(page1))
+        page1 = os.path.join(workspace.directory, 'OCR-D-OCR-CALAMARI/OCR-D-OCR-CALAMARI_0001.xml')
+        self.assertTrue(os.path.exists(page1))
         with open(page1, 'r') as f:
             self.assertIn('verſchuldeten', f.read())