init commit with face recgnition pipeline #2293

leo-q8 · 2024-10-21T15:36:13Z

No description provided.

paddle-bot · 2024-10-21T15:36:20Z

Thanks for your contribution!

Sunting78 · 2024-10-22T02:49:58Z

paddlex/configs/face_detection/BlazeFace.yaml

+  mode: check_dataset # check_dataset/train/evaluate/predict
+  dataset_dir: "/paddle/dataset/paddlex/det/widerface_coco_examples"
+  device: gpu:0,1,2,3
+  output: "output/blazeface"


统一为output吧文档默认路径都是output

Sunting78 · 2024-10-22T02:58:09Z

paddlex/configs/face_recognition/FaceRecognition_IResNet50_ArcFace.yaml

+Global:
+  model: IResNet50
+  mode: check_dataset # check_dataset/train/evaluate/predict
+  dataset_dir: "/paddle/dataset/paddlex/cls/cls_flowers_examples"


人脸识别数据格式和这个不一样吧，建一个人脸识别demo数据集

Sunting78 · 2024-10-22T02:58:56Z

paddlex/configs/face_recognition/FaceRecognition_IResNet50_ArcFace.yaml

+  log_interval: 1
+
+Export:
+  weight_path: https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/foundation_models/CLIP_vit_base_patch16_224_ft_in1k_pretrained.pdparams 


预训练模型是CLIP不对吧，这个换成你训练好的这个算法的人脸识别模型

Sunting78 · 2024-10-22T02:59:17Z

paddlex/configs/face_recognition/FaceRecognition_IResNet50_ArcFace.yaml

+    val_percent: null
+
+Train:
+  num_classes: 93431


类别数量需要和你上面建立的数据集的类别数量一致

Sunting78 · 2024-10-22T03:00:07Z

paddlex/configs/face_recognition/FaceRecognition_IResNet50_ArcFace.yaml

@@ -0,0 +1,41 @@
+Global:
+  model: IResNet50


算法名称需要和配置文件的文件名一致。文件名就叫做 IResNet50.yaml即可

Sunting78 · 2024-10-22T03:09:50Z

paddlex/repo_apis/PaddleClas_api/configs/FaceRecognition_MobileFaceNet_ArcFace.yaml

+      batch_size: 128
+      drop_last: False
+      shuffle: True
+      #sample_method: "id_avg_prob"


注释不要的删除

Sunting78 · 2024-10-22T03:10:09Z

paddlex/repo_apis/PaddleClas_api/configs/FaceRecognition_MobileFaceNet_ArcFace.yaml

+      num_workers: 4
+      use_shared_memory: True
+
+# Infer:


infer是需要配置的不然导出有问题

Sunting78 · 2024-10-22T03:13:56Z

paddlex/inference/utils/official_models.py

@@ -258,6 +258,7 @@
    "RT-DETR-H_layout_3cls": "https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0b1_v2/RT-DETR-H_layout_3cls_infer.tar",
    "RT-DETR-H_layout_17cls": "https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0b1_v2/RT-DETR-H_layout_17cls_infer.tar",
    "PicoDet_LCNet_x2_5_face": "https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0b1_v2/PicoDet_LCNet_x2_5_face_infer.tar",
+    "BlazeFace": "https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0b1_v2/BlazeFace_infer.tar" 


识别的模型也需要填写上

Sunting78 · 2024-10-22T03:18:34Z

paddlex/inference/pipelines/face_recognition/face_recognition.py

+# limitations under the License.
+
+import numpy as np
+from paddlex.inference.components import CropByBoxes, FaissIndexer


都改成相对的吧

Sunting78 · 2024-10-22T03:21:15Z

paddlex/inference/models/face_recognition.py

+import numpy as np
+
+from paddlex.utils.func_register import FuncRegister
+from paddlex.modules.face_recognition.model_list import MODELS


相对路径比如 from ...utils.func_register import FuncRegister

Sunting78 · 2024-10-22T03:32:48Z

paddlex/inference/components/retrieval/faiss.py

+            dtype = (
+                np.uint8 if self._metric_type in self.BINARY_METRIC_TYPE else np.float32
+            )
+            features = np.array(features).astype(dtype)


是否需要取平均呢？

Sunting78 · 2024-10-22T03:33:27Z

paddlex/inference/components/retrieval/faiss.py

+            pickle.dump(ids, fd)
+
+
+def get_file_list(data_file, root_dir, delimiter="\t"):


建库的文件组织方式和这个是不同的吧

Sunting78 · 2024-10-22T14:37:07Z

paddlex/inference/pipelines/face_recognition/face_recognition.py

+        super().__init__(device, predictor_kwargs)
+        self._build_predictor(det_model, rec_model)
+        self.set_predictor(det_batch_size, rec_batch_size, device)
+        self._indexer = (


这里加个is_build的参数吧，如果is_build=True 就调用self.build_index()建库，库保存在index_dir，如果是false相当于用户有自己的库，就传入index_dir就可以。

Sunting78 · 2024-10-22T14:42:28Z

paddlex/inference/pipelines/face_recognition/face_recognition.py

+        det_batch_size=1,
+        rec_batch_size=1,
+        index_dir=None,
+        metric_type="IP",


默认是L2还是IP？

cuicheng01 · 2024-10-23T02:44:27Z

paddlex/configs/face_detection/BlazeFace.yaml

+  batch_size: 4
+  learning_rate: 0.001
+  pretrain_weight_path: null
+  warmup_steps: 500


一共1000iter，warmup 500？

cuicheng01 · 2024-10-23T02:45:03Z

paddlex/configs/face_recognition/IResNet50.yaml

@@ -0,0 +1,41 @@
+Global:


这个模型要下掉吧？

cuicheng01 · 2024-10-23T02:45:35Z

paddlex/configs/face_recognition/MobileFaceNet.yaml

+CheckDataset:
+  convert: 
+    enable: False
+    src_dataset_ype: null


cuicheng01 · 2024-10-23T02:48:33Z

paddlex/modules/face_recognition/model_list.py

+
+MODELS = [
+    "MobileFaceNet",
+    "IResnet50"


这里名字又没对上

cuicheng01 · 2024-10-23T02:49:15Z

paddlex/modules/face_recognition/trainer.py

+                self.train_config.pretrain_weight_path
+            )
+
+        label_dict_path = Path(self.global_config.dataset_dir).joinpath("label.txt")


你的数据集还是否需要label.txt

cuicheng01 · 2024-10-23T02:50:01Z

paddlex/repo_apis/PaddleClas_api/configs/IResNet50.yaml

+  name: Momentum
+  momentum: 0.9
+  lr:
+    name: Piecewise


建议直接cosine梭哈

cuicheng01 · 2024-10-23T02:50:14Z

paddlex/repo_apis/PaddleClas_api/configs/IResNet50.yaml

+  Train:
+    dataset:
+      name: ImageNetDataset
+      image_root: dataset/MS1M_v3/


这里的root不对吧

cuicheng01 · 2024-10-23T02:51:33Z

paddlex/repo_apis/PaddleClas_api/configs/IResNet50.yaml

+  epochs: 25
+  print_batch_step: 20
+  use_visualdl: False
+  eval_mode: face_recognition


下边这些参数不应该在global中，应该放到Eval中

Sunting78 · 2024-10-23T06:39:50Z

paddlex/configs/face_recognition/IResNet50.yaml

+Global:
+  model: IResNet50
+  mode: check_dataset # check_dataset/train/evaluate/predict
+  dataset_dir: "/paddle/dataset/paddlex/cls/face_train_examples"


face_rec_examples

Sunting78 reviewed Oct 22, 2024

View reviewed changes

leo-q8 force-pushed the face_rec branch 4 times, most recently from f7a1c02 to acaaa64 Compare October 22, 2024 14:02

Sunting78 reviewed Oct 22, 2024

View reviewed changes

cuicheng01 reviewed Oct 23, 2024

View reviewed changes

Sunting78 reviewed Oct 23, 2024

View reviewed changes

init commit with face recgnition pipeline

16b8a1d

leo-q8 force-pushed the face_rec branch from acaaa64 to 16b8a1d Compare October 23, 2024 12:23

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

init commit with face recgnition pipeline #2293

init commit with face recgnition pipeline #2293

leo-q8 commented Oct 21, 2024

paddle-bot bot commented Oct 21, 2024

Sunting78 Oct 22, 2024

Sunting78 Oct 22, 2024

Sunting78 Oct 22, 2024

Sunting78 Oct 22, 2024

Sunting78 Oct 22, 2024

Sunting78 Oct 22, 2024

Sunting78 Oct 22, 2024

Sunting78 Oct 22, 2024

Sunting78 Oct 22, 2024

Sunting78 Oct 22, 2024

Sunting78 Oct 22, 2024

Sunting78 Oct 22, 2024

Sunting78 Oct 22, 2024

Sunting78 Oct 22, 2024

cuicheng01 Oct 23, 2024

cuicheng01 Oct 23, 2024

cuicheng01 Oct 23, 2024

cuicheng01 Oct 23, 2024

cuicheng01 Oct 23, 2024

cuicheng01 Oct 23, 2024

cuicheng01 Oct 23, 2024

cuicheng01 Oct 23, 2024

Sunting78 Oct 23, 2024

		pickle.dump(ids, fd)


		def get_file_list(data_file, root_dir, delimiter="\t"):

init commit with face recgnition pipeline #2293

Are you sure you want to change the base?

init commit with face recgnition pipeline #2293

Conversation

leo-q8 commented Oct 21, 2024

paddle-bot bot commented Oct 21, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment