fix: adjust batch sizes and improve performance settings in various modules

mineru/backend/pipeline/batch_analyze.py

@@ -12,6 +12,7 @@ from ...utils.ocr_utils import get_adjusted_mfdetrec_res, get_ocr_result_list, O
 YOLO_LAYOUT_BASE_BATCH_SIZE = 8
 MFD_BASE_BATCH_SIZE = 1
 MFR_BASE_BATCH_SIZE = 16
+OCR_DET_BASE_BATCH_SIZE = 16
 
 
 class BatchAnalyze:
@@ -170,9 +171,9 @@ class BatchAnalyze:
                         batch_images.append(padded_img)
 
                     # 批处理检测
-                    batch_size = min(len(batch_images), self.batch_ratio * 16)  # 增加批处理大小
-                    # logger.debug(f"OCR-det batch: {batch_size} images, target size: {target_h}x{target_w}")
-                    batch_results = ocr_model.text_detector.batch_predict(batch_images, batch_size)
+                    det_batch_size = min(len(batch_images), self.batch_ratio * OCR_DET_BASE_BATCH_SIZE)  # 增加批处理大小
+                    # logger.debug(f"OCR-det batch: {det_batch_size} images, target size: {target_h}x{target_w}")
+                    batch_results = ocr_model.text_detector.batch_predict(batch_images, det_batch_size)
 
                     # 处理批处理结果
                     for i, (crop_info, (dt_boxes, elapse)) in enumerate(zip(group_crops, batch_results)):

mineru/backend/pipeline/pipeline_analyze.py

@@ -74,10 +74,10 @@ def doc_analyze(
         table_enable=True,
 ):
     """
-    适当调大MIN_BATCH_INFERENCE_SIZE可以提高性能，可能会增加显存使用量，
-    可通过环境变量MINERU_MIN_BATCH_INFERENCE_SIZE设置，默认值为128。
+    适当调大MIN_BATCH_INFERENCE_SIZE可以提高性能，更大的 MIN_BATCH_INFERENCE_SIZE会消耗更多内存，
+    可通过环境变量MINERU_MIN_BATCH_INFERENCE_SIZE设置，默认值为384。
     """
-    min_batch_inference_size = int(os.environ.get('MINERU_MIN_BATCH_INFERENCE_SIZE', 128))
+    min_batch_inference_size = int(os.environ.get('MINERU_MIN_BATCH_INFERENCE_SIZE', 384))
 
     # 收集所有页面信息
     all_pages_info = []  # 存储(dataset_index, page_index, img, ocr, lang, width, height)

mineru/model/mfr/unimernet/Unimernet.py

@@ -115,7 +115,7 @@ class UnimernetModel(object):
                 mf_img = mf_img.to(dtype=self.model.dtype)
                 mf_img = mf_img.to(self.device)
                 with torch.no_grad():
-                    output = self.model.generate({"image": mf_img})
+                    output = self.model.generate({"image": mf_img}, batch_size=batch_size)
                 mfr_res.extend(output["fixed_str"])
 
                 # 更新进度条，每次增加batch_size，但要注意最后一个batch可能不足batch_size

mineru/model/mfr/unimernet/unimernet_hf/modeling_unimernet.py

@@ -468,7 +468,7 @@ class UnimernetModel(VisionEncoderDecoderModel):
         ).loss
         return {"loss": loss}
 
-    def generate(self, samples, do_sample: bool = False, temperature: float = 0.2, top_p: float = 0.95):
+    def generate(self, samples, do_sample: bool = False, temperature: float = 0.2, top_p: float = 0.95, batch_size=64):
         pixel_values = samples["image"]
         num_channels = pixel_values.shape[1]
         if num_channels == 1:
@@ -478,7 +478,13 @@ class UnimernetModel(VisionEncoderDecoderModel):
         if do_sample:
             kwargs["temperature"] = temperature
             kwargs["top_p"] = top_p
-        
+
+        if self.tokenizer.tokenizer.model_max_length > 1152:
+            if batch_size <= 32:
+                self.tokenizer.tokenizer.model_max_length = 1152  # 6g
+            else:
+                self.tokenizer.tokenizer.model_max_length = 1344  # 8g
+
         outputs = super().generate(
             pixel_values=pixel_values,
             max_new_tokens=self.tokenizer.tokenizer.model_max_length, # required

mineru/model/ocr/paddleocr2pytorch/pytorch_paddle.py

@@ -88,7 +88,7 @@ class PytorchPaddleOCR(TextSystem):
         kwargs['det_model_path'] = det_model_path
         kwargs['rec_model_path'] = rec_model_path
         kwargs['rec_char_dict_path'] = os.path.join(root_dir, 'pytorchocr', 'utils', 'resources', 'dict', dict_file)
-        # kwargs['rec_batch_num'] = 8
+        kwargs['rec_batch_num'] = 16
 
         kwargs['device'] = device