Fix the two FP16 models of CRNN(#131)

* The results of the two models remain the same accuracy.

* re-upload int8 models

* updated by yiyao

* remove the sim suffix

---------

Co-authored-by: zihaomu <[email protected]>

models/text_recognition_crnn/README.md

@@ -11,7 +11,7 @@ Results of accuracy evaluation with [tools/eval](../../tools/eval) at different
 | CRNN_EN_INT8 | 81.75      | 75.33     | 52.43     |
 | CRNN_CH      | 71.28      | 80.90     | 67.36     |
 | CRNN_CH_FP16 | 78.63      | 80.93     | 67.01     |
-| CRNN_CH_INT8 | 78.03      | 80.93     | 66.67     |
+| CRNN_CH_INT8 | 78.11      | 81.20     | 67.01     |
 
 \*: 'FP16' or 'INT8' stands for 'model quantized into FP16' or 'model quantized into int8'
 

tools/eval/datasets/icdar.py

@@ -41,7 +41,7 @@ class ICDAR:
 
             rbbox = np.array([0, img.shape[0], 0, 0, img.shape[1], 0, img.shape[1], img.shape[0]])
             pred = model.infer(img, rbbox)
-            if label == pred:
+            if label.lower() == pred.lower():
                 right_num += 1
 
         self.acc = right_num/(len(self.val_label) * 1.0)