Renaming model files to have more information on architecture, training data and more (#7)

fengyuentau · web-flow · commit 6a66e0d6e47a · 2021-11-01T16:50:09.000+08:00
* add suffix of training dataset, arch &amp; upload time to each model

* update DB-IC15 benchmark results
diff --git a/README.md b/README.md
@@ -5,6 +5,12 @@ A zoo for models tuned for OpenCV DNN with benchmarks on different platforms.
 Guidelines:
 - To clone this repo, please install [git-lfs](https://git-lfs.github.com/), run `git lfs install` and use `git lfs clone https://github.com/opencv/opencv_zoo`.
 - To run benchmark on your hardware settings, please refer to [benchmark/README](./benchmark/README.md).
+- Understand model filename: `<topic>_<model_name>_<dataset>_<arch>_<upload_time>`
+    - `<topic>`: research topics, such as `face detection` etc.
+    - `<model_name>`: exact model names.
+    - `<dataset>`: (Optional) the dataset that the model is trained with.
+    - `<arch>`: (Optional) the backbone architecture of the model.
+    - `<upload_time>`: the time when the model is uploaded, meaning the latest version of this model unless specified.
 
 ## Models & Benchmarks
 
@@ -16,19 +22,19 @@ Hardware Setup:
 ***Important Notes***:
 - The time data that shown on the following table presents the time elapsed from preprocess (resize is excluded), to a forward pass of a network, and postprocess to get final results.
 - The time data that shown on the following table is the median of 10 runs. Different metrics may be applied to some specific models.
+- Batch size is 1 for all benchmark results.
 - View [benchmark/config](./benchmark/config) for more details on benchmarking different models.
 
-| Model | Input Size | CPU x86_64 (ms) | CPU ARM (ms) | GPU CUDA (ms)* |
+| Model | Input Size | CPU x86_64 (ms) | CPU ARM (ms) | GPU CUDA (ms) |
 |-------|------------|-----------------|--------------|---------------|
 | [YuNet](./models/face_detection_yunet)   | 160x120 | 1.45   | 6.22    | 12.18 |
-| [DB](./models/text_detection_db)         | 640x480 | 137.38 | 2780.78 | 220.52 |
+| [DB-IC15](./models/text_detection_db)    | 640x480 | 142.91 | 2835.91 | 208.41 |
+| [DB-TD500](./models/text_detection_db)   | 640x480 | 142.91 | 2841.71 | 210.51 |
 | [CRNN](./models/text_recognition_crnn)   | 100x32  | 50.21  | 234.32  | 196.15 |
 | [SFace](./models/face_recognition_sface) | 112x112 | 8.65 | 99.20 | 24.88 |
 | [PP-ResNet](./models/image_classification_ppresnet) | 224x224 | 56.05 | 602.58 | 98.64 |
 | [PP-HumanSeg](./models/human_segmentation_pphumanseg) | 192x192 | 19.92 | 105.32 | 67.97 |
 
-*: Batch size is 1.
-
 ## License
 
 OpenCV Zoo is licensed under the [Apache 2.0 license](./LICENSE). Please refer to licenses of different models.
diff --git a/benchmark/config/face_detection_yunet.yaml b/benchmark/config/face_detection_yunet.yaml
@@ -16,7 +16,7 @@ Benchmark:
 
 Model:
   name: "YuNet"
-  modelPath: "models/face_detection_yunet/face_detection_yunet.onnx"
+  modelPath: "models/face_detection_yunet/face_detection_yunet_2021sep.onnx"
   confThreshold: 0.6
   nmsThreshold: 0.3
   topK: 5000
diff --git a/benchmark/config/face_recognition_sface.yaml b/benchmark/config/face_recognition_sface.yaml
@@ -14,4 +14,4 @@ Benchmark:
 
 Model:
   name: "SFace"
-  modelPath: "models/face_recognition_sface/face_recognition_sface.onnx"
+  modelPath: "models/face_recognition_sface/face_recognition_sface_2021sep.onnx"
diff --git a/benchmark/config/human_segmentation_pphumanseg.yaml b/benchmark/config/human_segmentation_pphumanseg.yaml
@@ -15,4 +15,4 @@ Benchmark:
 
 Model:
   name: "PPHumanSeg"
-  modelPath: "models/human_segmentation_pphumanseg/human_segmentation_pphumanseg.onnx"
+  modelPath: "models/human_segmentation_pphumanseg/human_segmentation_pphumanseg_2021oct.onnx"
diff --git a/benchmark/config/image_classification_ppresnet.yaml b/benchmark/config/image_classification_ppresnet.yaml
@@ -16,5 +16,5 @@ Benchmark:
 
 Model:
   name: "PPResNet"
-  modelPath: "models/image_classification_ppresnet/image_classification_ppresnet50.onnx"
+  modelPath: "models/image_classification_ppresnet/image_classification_ppresnet50_2021oct.onnx"
   labelPath: "models/image_classification_ppresnet/imagenet_labels.txt"
diff --git a/benchmark/config/text_detection_db.yaml b/benchmark/config/text_detection_db.yaml
@@ -15,7 +15,7 @@ Benchmark:
 
 Model:
   name: "DB"
-  modelPath: "models/text_detection_db/text_detection_db.onnx"
+  modelPath: "models/text_detection_db/text_detection_DB_TD500_resnet18_2021sep.onnx"
   binaryThreshold: 0.3
   polygonThreshold: 0.5
   maxCandidates: 200
diff --git a/benchmark/config/text_recognition_crnn.yaml b/benchmark/config/text_recognition_crnn.yaml
@@ -14,4 +14,4 @@ Benchmark:
 
 Model:
   name: "CRNN"
-  modelPath: "models/text_recognition_crnn/text_recognition_crnn.onnx"
+  modelPath: "models/text_recognition_crnn/text_recognition_CRNN_VGG_BiLSTM_CTC_2021sep.onnx"
diff --git a/models/face_detection_yunet/README.md b/models/face_detection_yunet/README.md
@@ -2,6 +2,10 @@
 
 YuNet is a light-weight, fast and accurate face detection model, which achieves 0.834(AP_easy), 0.824(AP_medium), 0.708(AP_hard) on the WIDER Face validation set.
 
+Notes:
+- Model source: [here](https://github.com/ShiqiYu/libfacedetection.train/blob/a61a428929148171b488f024b5d6774f93cdbc13/tasks/task1/onnx/yunet.onnx).
+- For details on training this model, please visit https://github.com/ShiqiYu/libfacedetection.train.
+
 ## Demo
 
 Run the following command to try the demo:
diff --git a/models/face_detection_yunet/face_detection_yunet_2021sep.onnx b/models/face_detection_yunet/face_detection_yunet_2021sep.onnx
diff --git a/models/face_recognition_sface/README.md b/models/face_recognition_sface/README.md
@@ -2,15 +2,15 @@
 
 SFace: Sigmoid-Constrained Hypersphere Loss for Robust Face Recognition
 
-SFace is contributed by [Yaoyao Zhong](https://github.com/zhongyy/SFace). [face_recognition_sface.onnx](./face_recognition_sface.onnx) is converted from the model from https://github.com/zhongyy/SFace thanks to [Chengrui Wang](https://github.com/crywang).
-
 Note:
-- There is [a PR for OpenCV adding this model](https://github.com/opencv/opencv/pull/20422) to work with OpenCV DNN in C++ implementation.
-- Support 5-landmark warp for now.
-- `demo.py` requires [../face_detection_yunet](../face_detection_yunet) to run.
+- SFace is contributed by [Yaoyao Zhong](https://github.com/zhongyy/SFace).
+- [face_recognition_sface_2021sep.onnx](./face_recognition_sface_2021sep.onnx) is converted from the model from https://github.com/zhongyy/SFace thanks to [Chengrui Wang](https://github.com/crywang).
+- Support 5-landmark warpping for now (2021sep)
 
 ## Demo
 
+***NOTE***: This demo uses [../face_detection_yunet](../face_detection_yunet) as face detector, which supports 5-landmark detection for now (2021sep).
+
 Run the following command to try the demo:
 ```shell
 # recognize on images
diff --git a/models/face_recognition_sface/face_recognition_sface_2021sep.onnx b/models/face_recognition_sface/face_recognition_sface_2021sep.onnx
diff --git a/models/human_segmentation_pphumanseg/human_segmentation_pphumanseg_2021oct.onnx b/models/human_segmentation_pphumanseg/human_segmentation_pphumanseg_2021oct.onnx
diff --git a/models/image_classification_ppresnet/image_classification_ppresnet50_2021oct.onnx b/models/image_classification_ppresnet/image_classification_ppresnet50_2021oct.onnx
diff --git a/models/text_detection_db/README.md b/models/text_detection_db/README.md
@@ -2,7 +2,11 @@
 
 Real-time Scene Text Detection with Differentiable Binarization
 
-`text_detection_db.onnx` is trained on [TD500 dataset](http://www.iapr-tc11.org/mediawiki/index.php/MSRA_Text_Detection_500_Database_(MSRA-TD500)), which can detect both English & Chinese instances. It is obtained from [here](https://docs.opencv.org/master/d4/d43/tutorial_dnn_text_spotting.html) and renamed from `DB_TD500_resnet18.onnx`.
+Note:
+- Models source: [here](https://drive.google.com/drive/folders/1qzNCHfUJOS0NEUOIKn69eCtxdlNPpWbq).
+- `IC15` in the filename means the model is trained on [IC15 dataset](https://rrc.cvc.uab.es/?ch=4&com=introduction), which can detect English text instances only.
+- `TD500` in the filename means the model is trained on [TD500 dataset](http://www.iapr-tc11.org/mediawiki/index.php/MSRA_Text_Detection_500_Database_(MSRA-TD500)), which can detect both English & Chinese instances.
+- Visit https://docs.opencv.org/master/d4/d43/tutorial_dnn_text_spotting.html for more information.
 
 ## Demo
 
diff --git a/models/text_detection_db/demo.py b/models/text_detection_db/demo.py
@@ -21,7 +21,7 @@ def str2bool(v):
 
 parser = argparse.ArgumentParser(description='Real-time Scene Text Detection with Differentiable Binarization (https://arxiv.org/abs/1911.08947).')
 parser.add_argument('--input', '-i', type=str, help='Path to the input image. Omit for using default camera.')
-parser.add_argument('--model', '-m', type=str, default='text_detection_db.onnx', help='Path to the model.')
+parser.add_argument('--model', '-m', type=str, default='text_detection_DB_TD500_resnet18.onnx', help='Path to the model.')
 parser.add_argument('--width', type=int, default=736,
                     help='Preprocess input image by resizing to a specific width. It should be multiple by 32.')
 parser.add_argument('--height', type=int, default=736,
diff --git a/models/text_detection_db/text_detection_DB_IC15_resnet18_2021sep.onnx b/models/text_detection_db/text_detection_DB_IC15_resnet18_2021sep.onnx
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:62fce3a058ac519952062fbd2cf45bae661130ce753d2a0fc918f4776ba3a02a
+size 48770395
diff --git a/models/text_detection_db/text_detection_DB_TD500_resnet18_2021sep.onnx b/models/text_detection_db/text_detection_DB_TD500_resnet18_2021sep.onnx
diff --git a/models/text_recognition_crnn/README.md b/models/text_recognition_crnn/README.md
@@ -2,11 +2,13 @@
 
 An End-to-End Trainable Neural Network for Image-based Sequence Recognition and Its Application to Scene Text Recognition
 
-`text_recognition_crnn.onnx` is trained using the code from https://github.com/zihaomu/deep-text-recognition-benchmark, which can only recognize english words. It is obtained from https://drive.google.com/drive/folders/1cTbQ3nuZG-EKWak6emD_s8_hHXWz7lAr and renamed from `CRNN_VGG_BiLSTM_CTC.onnx`. Visit https://docs.opencv.org/4.5.2/d9/d1e/tutorial_dnn_OCR.html for more information.
+Note:
+- Model source: https://docs.opencv.org/4.5.2/d9/d1e/tutorial_dnn_OCR.html.
+- For details on training this model, please visit https://github.com/zihaomu/deep-text-recognition-benchmark, which can only recognize english words.
 
 ## Demo
 
-***NOTE**: This demo use [text_detection_db](../text_detection_db) as text detector.
+***NOTE***: This demo uses [text_detection_db](../text_detection_db) as text detector.
 
 Run the following command to try the demo:
 ```shell
diff --git a/models/text_recognition_crnn/demo.py b/models/text_recognition_crnn/demo.py
@@ -26,7 +26,7 @@ def str2bool(v):
 parser = argparse.ArgumentParser(
     description="An End-to-End Trainable Neural Network for Image-based Sequence Recognition and Its Application to Scene Text Recognition (https://arxiv.org/abs/1507.05717)")
 parser.add_argument('--input', '-i', type=str, help='Path to the input image. Omit for using default camera.')
-parser.add_argument('--model', '-m', type=str, default='text_recognition_crnn.onnx', help='Path to the model.')
+parser.add_argument('--model', '-m', type=str, default='text_recognition_CRNN_VGG_BiLSTM_CTC.onnx', help='Path to the model.')
 parser.add_argument('--width', type=int, default=736,
                     help='The width of input image being sent to the text detector.')
 parser.add_argument('--height', type=int, default=736,
diff --git a/models/text_recognition_crnn/text_recognition_CRNN_VGG_BiLSTM_CTC_2021sep.onnx b/models/text_recognition_crnn/text_recognition_CRNN_VGG_BiLSTM_CTC_2021sep.onnx
