Hommoner

in my experience pip install paddlepaddle-gpu==2.3.0.post112 pip install paddlepaddle-gpu==2.2.0.post112 are good in paddle detection-> maskrcnn others may not working

in demo.py line 145 clss = ins.pred_classes.cpu().numpy()

reference to ./tools/compute_anchors.py you can change some code like below run in your IDE then get Final anchor 

I found the leak is due to every time we call this line "opt = Optimizer(input_tensor, losses, wrt_tensor=penultimate_output, norm_grads=False)" , the tensorflow graph will add new tensor. My workaround is...

code like this: create init once function and generate map function def visualize_cam_init(model, layer_idx, filter_indices): penultimate_layer = _find_penultimate_layer(model, layer_idx, None) losses = [ (ActivationMaximization(model.layers[layer_idx], filter_indices), -1) ] penultimate_output = penultimate_layer.output...

> @Hommoner Thank for sharing the code. > > Why you define several opt, rather than just use one. opt is the same for every class_id, right? Or you want...

試試修改源代碼 ``` if os.path.exists('./data/paddle.png'): result = self.table_ocr('./data/paddle.png', return_ocr_result_in_table=True) ``` 為 ``` if os.path.exists('./data/paddle.png'): result = self.ocr.ocr('./data/paddle.png', cls=True, det=True) ``` 或是直接拿掉

試試 #11459

試試修改PPOCRLabel.py ``` def modelChoose(self): print(self.comboBox.currentText()) lg_idx = {'Chinese & English': 'ch', 'English': 'en', 'French': 'french', 'German': 'german', 'Korean': 'korean', 'Japanese': 'japan'} del self.ocr self.ocr = PaddleOCR(use_pdserving=False, use_angle_cls=True, det=True, cls=True, use_gpu=False,...

.h檔定義 ``` fastdeploy::pipeline::PPOCRv3* ppocr_v3 = NULL; fastdeploy::vision::ocr::DBDetector* det_model = NULL; fastdeploy::vision::ocr::Classifier* cls_model = NULL; fastdeploy::vision::ocr::Recognizer* rec_model = NULL; ``` .cpp檔子程序初始化類似這樣 ``` if (rec_model) { delete rec_model; rec_model = NULL; }...