Support batch_size>1

dan/ocr/decoder.py

@@ -527,7 +527,7 @@ class CTCLanguageDecoder:
         ctc_probs = ctc_probs.log()
 
         # Insert CTC tensor between frames
-        for fn in range(batch_frames[0] - 1):
+        for fn in range(batch_frames.max() - 1):
             batch_features = torch.cat(
                 [
                     batch_features[:, : 2 * fn + 1, :],

dan/ocr/predict/prediction.py

@@ -508,7 +508,6 @@ def run(
     dan_model.load(
         model, parameters, charset, mode="eval", use_language_model=use_language_model
     )
-    batch_size = 1 if use_language_model else batch_size
 
     images = image_dir.rglob(f"*{image_extension}") if not image else [image]
     for image_batch in list_to_batches(images, n=batch_size):