merge dygraph

test1/table/README_ch.md

@@ -19,7 +19,29 @@
 
 
 ### 2.1 训练
-TBD
+#### 数据准备  
+训练数据使用公开数据集[PubTabNet](https://arxiv.org/abs/1911.10683)，可以从[官网](https://github.com/ibm-aur-nlp/PubTabNet)下载。PubTabNet数据集包含约50万张表格数据的图像，以及图像对应的html格式的注释。
+
+#### 启动训练  
+*如果您安装的是cpu版本，请将配置文件中的 `use_gpu` 字段修改为false*
+```shell
+# 单机单卡训练
+python3 tools/train.py -c configs/table/table_mv3.yml
+# 单机多卡训练，通过 --gpus 参数设置使用的GPU ID
+python3 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/table/table_mv3.yml
+```
+
+上述指令中，通过-c 选择训练使用configs/table/table_mv3.yml配置文件。有关配置文件的详细解释，请参考[链接](./config.md)。
+
+#### 断点训练
+
+如果训练程序中断，如果希望加载训练中断的模型从而恢复训练，可以通过指定Global.checkpoints指定要加载的模型路径：
+```shell
+python3 tools/train.py -c configs/table/table_mv3.yml -o Global.checkpoints=./your/trained/model
+```
+
+**注意**：`Global.checkpoints`的优先级高于`Global.pretrain_weights`的优先级，即同时指定两个参数时，优先加载`Global.checkpoints`指定的模型，如果`Global.checkpoints`指定的模型路径有误，会加载`Global.pretrain_weights`指定的模型。
+
 
 ### 2.2 评估
 先cd到PaddleOCR/ppstructure目录下

tools/eval.py

@@ -27,7 +27,7 @@ from ppocr.data import build_dataloader
 from ppocr.modeling.architectures import build_model
 from ppocr.postprocess import build_post_process
 from ppocr.metrics import build_metric
-from ppocr.utils.save_load import init_model
+from ppocr.utils.save_load import init_model, load_pretrained_params
 from ppocr.utils.utility import print_dict
 import tools.program as program
 
@@ -55,7 +55,10 @@ def main():
 
     model = build_model(config['Architecture'])
     use_srn = config['Architecture']['algorithm'] == "SRN"
-    model_type = config['Architecture']['model_type']
+    if "model_type" in config['Architecture'].keys():
+        model_type = config['Architecture']['model_type']
+    else:
+        model_type = None
 
     best_model_dict = init_model(config, model)
     if len(best_model_dict):
@@ -68,7 +71,7 @@ def main():
 
     # start eval
     metric = program.eval(model, valid_dataloader, post_process_class,
-                          eval_class, model_type, use_srn)
+                        eval_class, model_type, use_srn)
     logger.info('metric eval ***************')
     for k, v in metric.items():
         logger.info('{}:{}'.format(k, v))

tools/infer/predict_cls.py

@@ -112,7 +112,6 @@ class TextClassifier(object):
                 if '180' in label and score > self.cls_thresh:
                     img_list[indices[beg_img_no + rno]] = cv2.rotate(
                         img_list[indices[beg_img_no + rno]], 1)
-        elapse = time.time() - starttime
         return img_list, cls_res, elapse
 
 
@@ -146,7 +145,6 @@ def main(args):
                                                cls_res[ino]))
     logger.info(
         "The predict time about text angle classify module is as follows: ")
-    text_classifier.cls_times.info(average=False)
 
 
 if __name__ == "__main__":

tools/infer/predict_det.py

@@ -175,7 +175,7 @@ class TextDetector(object):
 
         st = time.time()
 
-        if args.benchmark:
+        if self.args.benchmark:
             self.autolog.times.start()
 
         data = transform(data, self.preprocess_op)
@@ -186,7 +186,7 @@ class TextDetector(object):
         shape_list = np.expand_dims(shape_list, axis=0)
         img = img.copy()
 
-        if args.benchmark:
+        if self.args.benchmark:
             self.autolog.times.stamp()
 
         self.input_tensor.copy_from_cpu(img)
@@ -195,7 +195,7 @@ class TextDetector(object):
         for output_tensor in self.output_tensors:
             output = output_tensor.copy_to_cpu()
             outputs.append(output)
-        if args.benchmark:
+        if self.args.benchmark:
             self.autolog.times.stamp()
 
         preds = {}
@@ -220,7 +220,7 @@ class TextDetector(object):
         else:
             dt_boxes = self.filter_tag_det_res(dt_boxes, ori_im.shape)
 
-        if args.benchmark:
+        if self.args.benchmark:
             self.autolog.times.end(stamp=True)
         et = time.time()
         return dt_boxes, et - st

tools/infer/predict_rec.py

@@ -64,6 +64,24 @@ class TextRecognizer(object):
         self.postprocess_op = build_post_process(postprocess_params)
         self.predictor, self.input_tensor, self.output_tensors, self.config = \
             utility.create_predictor(args, 'rec', logger)
+        self.benchmark = args.benchmark
+        if args.benchmark:
+            import auto_log
+            pid = os.getpid()
+            self.autolog = auto_log.AutoLogger(
+                model_name="rec",
+                model_precision=args.precision,
+                batch_size=args.rec_batch_num,
+                data_shape="dynamic",
+                save_path=args.save_log_path,
+                inference_config=self.config,
+                pids=pid,
+                process_name=None,
+                gpu_ids=0 if args.use_gpu else None,
+                time_keys=[
+                    'preprocess_time', 'inference_time', 'postprocess_time'
+                ],
+                warmup=10)
 
     def resize_norm_img(self, img, max_wh_ratio):
         imgC, imgH, imgW = self.rec_image_shape
@@ -168,6 +186,8 @@ class TextRecognizer(object):
         rec_res = [['', 0.0]] * img_num
         batch_num = self.rec_batch_num
         st = time.time()
+        if self.benchmark:
+            self.autolog.times.start()
         for beg_img_no in range(0, img_num, batch_num):
             end_img_no = min(img_num, beg_img_no + batch_num)
             norm_img_batch = []
@@ -196,6 +216,8 @@ class TextRecognizer(object):
                     norm_img_batch.append(norm_img[0])
             norm_img_batch = np.concatenate(norm_img_batch)
             norm_img_batch = norm_img_batch.copy()
+            if self.benchmark:
+                self.autolog.times.stamp()
 
             if self.rec_algorithm == "SRN":
                 encoder_word_pos_list = np.concatenate(encoder_word_pos_list)
@@ -222,6 +244,8 @@ class TextRecognizer(object):
                 for output_tensor in self.output_tensors:
                     output = output_tensor.copy_to_cpu()
                     outputs.append(output)
+                if self.benchmark:
+                    self.autolog.times.stamp()
                 preds = {"predict": outputs[2]}
             else:
                 self.input_tensor.copy_from_cpu(norm_img_batch)
@@ -231,11 +255,14 @@ class TextRecognizer(object):
                 for output_tensor in self.output_tensors:
                     output = output_tensor.copy_to_cpu()
                     outputs.append(output)
+                if self.benchmark:
+                    self.autolog.times.stamp()
                 preds = outputs[0]
             rec_result = self.postprocess_op(preds)
             for rno in range(len(rec_result)):
                 rec_res[indices[beg_img_no + rno]] = rec_result[rno]
-
+            if self.benchmark:
+                self.autolog.times.end(stamp=True)
         return rec_res, time.time() - st
 
 
@@ -251,9 +278,6 @@ def main(args):
         for i in range(10):
             res = text_recognizer([img])
 
-    cpu_mem, gpu_mem, gpu_util = 0, 0, 0
-    count = 0
-
     for image_file in image_file_list:
         img, flag = check_and_read_gif(image_file)
         if not flag:
@@ -273,6 +297,8 @@ def main(args):
     for ino in range(len(img_list)):
         logger.info("Predicts of {}:{}".format(valid_image_file_list[ino],
                                                rec_res[ino]))
+    if args.benchmark:
+        text_recognizer.autolog.report()
 
 
 if __name__ == "__main__":

tools/infer/predict_system.py

@@ -174,8 +174,6 @@ def main(args):
     logger.info("The predict total time is {}".format(time.time() - _st))
     logger.info("\nThe predict total time is {}".format(total_time))
 
-    img_num = text_sys.text_detector.det_times.img_num
-
 
 if __name__ == "__main__":
     args = utility.parse_args()

tools/infer/utility.py

@@ -34,7 +34,7 @@ def init_args():
     parser.add_argument("--use_gpu", type=str2bool, default=True)
     parser.add_argument("--ir_optim", type=str2bool, default=True)
     parser.add_argument("--use_tensorrt", type=str2bool, default=False)
-    parser.add_argument("--min_subgraph_size", type=int, default=3)
+    parser.add_argument("--min_subgraph_size", type=int, default=10)
     parser.add_argument("--precision", type=str, default="fp32")
     parser.add_argument("--gpu_mem", type=int, default=500)
 
@@ -161,7 +161,7 @@ def create_predictor(args, mode, logger):
         config.enable_use_gpu(args.gpu_mem, 0)
         if args.use_tensorrt:
             config.enable_tensorrt_engine(
-                precision_mode=inference.PrecisionType.Float32,
+                precision_mode=precision,
                 max_batch_size=args.max_batch_size,
                 min_subgraph_size=args.min_subgraph_size)
             # skip the minmum trt subgraph

tools/program.py

@@ -186,7 +186,10 @@ def train(config,
     model.train()
 
     use_srn = config['Architecture']['algorithm'] == "SRN"
-    model_type = config['Architecture']['model_type']
+    try: 
+        model_type = config['Architecture']['model_type']
+    except: 
+        model_type = None
 
     if 'start_epoch' in best_model_dict:
         start_epoch = best_model_dict['start_epoch']

tools/train.py

@@ -98,7 +98,6 @@ def main(config, device, logger, vdl_writer):
     eval_class = build_metric(config['Metric'])
     # load pretrain model
     pre_best_model_dict = load_dygraph_params(config, model, logger, optimizer)
-
     logger.info('train dataloader has {} iters'.format(len(train_dataloader)))
     if valid_dataloader is not None:
         logger.info('valid dataloader has {} iters'.format(