Fixes uploaded model being created with the wrong tag, disables block masking...

Fixes uploaded model being created with the wrong tag, disables block masking for now, freezes numpy version.

app/apps/core/forms.py

@@ -120,6 +120,8 @@ class DocumentProcessForm(BootstrapFormMixin, forms.Form):
     )
     segmentation_steps = forms.ChoiceField(choices=SEGMENTATION_STEPS_CHOICES,
                                            initial='lines', required=False)
+    seg_model = forms.ModelChoiceField(queryset=OcrModel.objects.filter(job=OcrModel.MODEL_JOB_SEGMENT),
+                                       label=_("Model"), required=False)
     override = forms.BooleanField(required=False, initial=False,
                                   help_text=_("If checked, deletes existing segmentation <b>and bound transcriptions</b> first!"))
     TEXT_DIRECTION_CHOICES = (('horizontal-lr', _("Horizontal l2r")),
@@ -142,8 +144,6 @@ class DocumentProcessForm(BootstrapFormMixin, forms.Form):
     transcription = forms.ModelChoiceField(queryset=Transcription.objects.all(), required=False)
 
     # segtrain
-    seg_model = forms.ModelChoiceField(queryset=OcrModel.objects.filter(job=OcrModel.MODEL_JOB_SEGMENT),
-                                       label=_("Model"), required=False)
     segtrain_model = forms.ModelChoiceField(queryset=OcrModel.objects.filter(job=OcrModel.MODEL_JOB_SEGMENT),
                                             label=_("Model"), required=False)
     
@@ -162,6 +162,7 @@ class DocumentProcessForm(BootstrapFormMixin, forms.Form):
         self.fields['binarizer'].widget.attrs['disabled'] = True
         self.fields['train_model'].queryset &= OcrModel.objects.filter(document=self.document)
         self.fields['segtrain_model'].queryset &= OcrModel.objects.filter(document=self.document)
+        self.fields['seg_model'].queryset &= OcrModel.objects.filter(document=self.document)
         self.fields['ocr_model'].queryset = OcrModel.objects.filter(
             Q(document=None, script=document.main_script)
             | Q(document=self.document))
@@ -200,7 +201,9 @@ class DocumentProcessForm(BootstrapFormMixin, forms.Form):
         data = super().clean()
         task = data.get('task')
 
-        if task == self.TASK_SEGTRAIN:
+        if task == self.TASK_SEGMENT:
+            model_job = OcrModel.MODEL_JOB_SEGMENT
+        elif task == self.TASK_SEGTRAIN:
             model_job = OcrModel.MODEL_JOB_SEGMENT
             if len(self.parts) < 2:
                 raise forms.ValidationError("Segmentation training requires at least 2 images.")

app/apps/core/models.py

@@ -583,27 +583,28 @@ class DocumentPart(OrderedModel):
                 options['model'] = model.file.path
             else:
                 options['model'] = settings.KRAKEN_DEFAULT_SEGMENTATION_MODEL
-            blocks = self.blocks.all()
-            if blocks:
-                for block in blocks:
-                    if block.box[2] < block.box[0] + 10 or block.box[3] < block.box[1] + 10:
-                        continue
-                    ic = im.crop(block.box)
-                    res = blla.segment(ic, **options)
-                    # if script_detect:
-                    #     res = pageseg.detect_scripts(im, res, valid_scripts=allowed_scripts)
-                    for line in res['lines']:
-                        Line.objects.create(
-                            document_part=self, block=block,
-                            box=(line[0]+block.box[0], line[1]+block.box[1],
-                                 line[2]+block.box[0], line[3]+block.box[1]))
-            else:
-                res = blla.segment(im, **options)
-                for line in res['lines']:
-                    newline = Line.objects.create(
-                        document_part=self,
-                        baseline=line['baseline'],
-                        mask=line['boundary'] if line['boundary'] is not None else None)
+            # blocks = self.blocks.all()
+            # if blocks:
+            #     for block in blocks:
+            #         if block.box[2] < block.box[0] + 10 or block.box[3] < block.box[1] + 10:
+            #             continue
+            #         ic = im.crop(block.box)
+            #         res = blla.segment(ic, **options)
+            #         # if script_detect:
+            #         #     res = pageseg.detect_scripts(im, res, valid_scripts=allowed_scripts)
+            #         for line in res['lines']:
+            #             Line.objects.create(
+            #                 document_part=self, block=block,
+            #                 box=(line[0]+block.box[0], line[1]+block.box[1],
+            #                      line[2]+block.box[0], line[3]+block.box[1]))
+            # else:
+            res = blla.segment(im, **options)
+            for line in res['lines']:
+                mask = line['boundary'] if line['boundary'] is not None else None
+                newline = Line.objects.create(
+                    document_part=self,
+                    baseline=line['baseline'],
+                    mask=mask)
         
         self.workflow_state = self.WORKFLOW_STATE_SEGMENTED
         self.save()

app/requirements.txt

@@ -20,3 +20,4 @@ drf-nested-routers==0.91
 bleach==3.1.0
 beautifulsoup4==4.7.1
 requests==2.21.0
+numpy>=1.17