add export_model doc (#3709)

* add export_model doc

README_cn.md

@@ -75,6 +75,7 @@ PaddleDetection的目的是为工业界和学术界提供丰富、易用的目
 
 ## 推理部署
 
+- [模型导出教程](docs/EXPORT_MODEL.md)
 - [C++推理部署](inference/README.md)
 
 ## Benchmark

docs/EXPORT_MODEL.md

+# 模型导出
+
+训练得到一个满足要求的模型后，如果想要将该模型接入到C++预测库或者Serving服务，需要通过`tools/export_model.py`导出该模型。
+
+## 启动参数说明
+
+|      FLAG      |      用途      |    默认值    |                 备注                      |
+|:--------------:|:--------------:|:------------:|:-----------------------------------------:|
+|       -c       |  指定配置文件  |     None     |                                           |
+|  --output_dir  |  模型保存路径  |  `./output`  |  模型默认保存在`output/配置文件名/`路径下 |
+
+## 使用示例
+
+使用[训练/评估/推断](GETTING_STARTED_cn.md)中训练得到的模型进行试用，脚本如下
+
+```bash
+# export model for RCNN
+python tools/export_model.py -c configs/faster_rcnn_r50_1x.yml \
+        --output_dir=./inference_model \
+        -o weights=output/faster_rcnn_r50_1x/model_final \
+           FasterRCNNTestFeed.image_shape=[3,800,1333]
+
+# export model for YOLOv3
+python tools/export_model.py -c configs/yolov3_darknet.yml \
+        --output_dir=./inference_model \
+        -o weights=output/yolov3_darknet/model_final \
+           YoloTestFeed.image_shape=[3,800,1333]
+
+# export model for SSD
+python tools/export_model.py -c configs/ssd/ssd_mobilenet_v1_voc.yml \
+        --output_dir=./inference_model \
+        -o weights=output/ssd_mobilenet_v1_voc/model_final \
+           SSDTestFeed.image_shape=[3,300,300]
+
+```
+
+- 预测模型会导出到`output/faster_rcnn_r50_1x`目录下，模型名和参数名分别为`__model__`和`__params__`。
+- 通过`image_shape`修改保存模型中的图片大小。使用Fluid-TensorRT进行预测时，由于TensorRT仅支持定长输入，需要将输入图片大小与`image_shape`保持一致。

docs/GETTING_STARTED.md

@@ -40,7 +40,6 @@ list below can be viewed by `--help`
 |       --json_eval        |       eval     |  Whether to evaluate with already existed bbox.json or mask.json  |  False  |  json path is set in `--output_eval`  |
 |       --output_dir       |      infer     |  Directory for storing the output visualization files  |  `./output`  |  `--output_dir output`  |
 |    --draw_threshold      |      infer     |  Threshold to reserve the result for visualization  |  0.5  |  `--draw_threshold 0.7`  |
-|  --save\_inference_model |      infer      |  Whether to save inference model in output_dir  |  False  |  save_inference_model is saved in `--output_dir`  |
 |      --infer_dir         |       infer     |  Directory for images to perform inference on  |  None  |    |
 |      --infer_img         |       infer     |  Image path  |  None  |  higher priority over --infer_dir  |
 |        --use_tb          |   train/infer   |  Whether to record the data with [tb-paddle](https://github.com/linshuliang/tb-paddle), so as to display in Tensorboard  |  False  |      |
@@ -149,17 +148,16 @@ moment, but it is a planned feature
   Different thresholds will produce different results depending on the calculation of [NMS](https://ieeexplore.ieee.org/document/1699659).
 
 
-- Save inference model
+- Export model
 
   ```bash
-  export CUDA_VISIBLE_DEVICES=0
-  python tools/infer.py -c configs/faster_rcnn_r50_1x.yml \
-                      --infer_img=demo/000000570688.jpg \
-                      --save_inference_model
+  python tools/export_model.py -c configs/faster_rcnn_r50_1x.yml \
+                      --output_dir=inference_model \
+                      -o weights=output/faster_rcnn_r50_1x/model_final \
+                         FasterRCNNTestFeed.image_shape=[3,800,1333]
   ```
 
-  Save inference model by set `--save_inference_model`, which can be loaded by PaddlePaddle predict library.
-
+  Save inference model `tools/export_model.py`, which can be loaded by PaddlePaddle predict library.
 
 ## FAQ
 

docs/GETTING_STARTED_cn.md

@@ -37,7 +37,6 @@ python tools/infer.py -c configs/faster_rcnn_r50_1x.yml --infer_img=demo/0000005
 |       --json_eval        |       eval     |  是否通过已存在的bbox.json或者mask.json进行评估  |  False  |  json文件路径在`--output_eval`中设置  |
 |       --output_dir       |      infer     |  输出推断后可视化文件  |  `./output`  |  `--output_dir output`  |
 |    --draw_threshold      |      infer     |  可视化时分数阈值  |  0.5  |  `--draw_threshold 0.7`  |
-|  --save\_inference_model |      infer      |  是否保存预测模型  |  False  |  预测模型保存路径为`--output_dir`设置的路径  |
 |      --infer_dir         |       infer     |  用于推断的图片文件夹路径  |  None  |    |
 |      --infer_img         |       infer     |  用于推断的图片路径  |  None  |  相较于`--infer_dir`具有更高优先级  |
 |        --use_tb          |   train/infer   |  是否使用[tb-paddle](https://github.com/linshuliang/tb-paddle)记录数据，进而在TensorBoard中显示  |  False  |      |
@@ -145,18 +144,6 @@ python -m paddle.distributed.launch --selected_gpus 0,1,2,3,4,5,6,7 tools/train.
   `--draw_threshold` 是个可选参数. 根据 [NMS](https://ieeexplore.ieee.org/document/1699659) 的计算，
   不同阈值会产生不同的结果。如果用户需要对自定义路径的模型进行推断，可以设置`-o weights`指定模型路径。
 
-- 保存推断模型
-
-  ```bash
-  # GPU推断
-  export CUDA_VISIBLE_DEVICES=0
-  python -u tools/infer.py -c configs/faster_rcnn_r50_1x.yml --infer_img=demo/000000570688.jpg \
-                      --save_inference_model
-  ```
-
-  通过设置`--save_inference_model`保存可供PaddlePaddle预测库加载的推断模型。
-
-
 ## FAQ
 
 **Q:**  为什么我使用单GPU训练loss会出`NaN`? </br>

tools/export_model.py

+# Copyright (c) 2019 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os
+
+from paddle import fluid
+
+from ppdet.core.workspace import load_config, merge_config, create
+from ppdet.modeling.model_input import create_feed
+from ppdet.utils.cli import ArgsParser
+from ppdet.utils.check import check_gpu
+import ppdet.utils.checkpoint as checkpoint
+
+import logging
+FORMAT = '%(asctime)s-%(levelname)s: %(message)s'
+logging.basicConfig(level=logging.INFO, format=FORMAT)
+logger = logging.getLogger(__name__)
+
+
+def prune_feed_vars(feeded_var_names, target_vars, prog):
+    """
+    Filter out feed variables which are not in program,
+    pruned feed variables are only used in post processing
+    on model output, which are not used in program, such
+    as im_id to identify image order, im_shape to clip bbox
+    in image.
+    """
+    exist_var_names = []
+    prog = prog.clone()
+    prog = prog._prune(targets=target_vars)
+    global_block = prog.global_block()
+    for name in feeded_var_names:
+        try:
+            v = global_block.var(name)
+            exist_var_names.append(str(v.name))
+        except Exception:
+            logger.info('save_inference_model pruned unused feed '
+                        'variables {}'.format(name))
+            pass
+    return exist_var_names
+
+
+def save_infer_model(FLAGS, exe, feed_vars, test_fetches, infer_prog):
+    cfg_name = os.path.basename(FLAGS.config).split('.')[0]
+    save_dir = os.path.join(FLAGS.output_dir, cfg_name)
+    feed_var_names = [var.name for var in feed_vars.values()]
+    target_vars = list(test_fetches.values())
+    feed_var_names = prune_feed_vars(feed_var_names, target_vars, infer_prog)
+    logger.info("Export inference model to {}, input: {}, output: "
+                "{}...".format(save_dir, feed_var_names,
+                               [str(var.name) for var in target_vars]))
+    fluid.io.save_inference_model(
+        save_dir,
+        feeded_var_names=feed_var_names,
+        target_vars=target_vars,
+        executor=exe,
+        main_program=infer_prog,
+        params_filename="__params__")
+
+
+def main():
+    cfg = load_config(FLAGS.config)
+
+    if 'architecture' in cfg:
+        main_arch = cfg.architecture
+    else:
+        raise ValueError("'architecture' not specified in config file.")
+
+    merge_config(FLAGS.opt)
+
+    if 'test_feed' not in cfg:
+        test_feed = create(main_arch + 'TestFeed')
+    else:
+        test_feed = create(cfg.test_feed)
+
+    # Use CPU for exporting inference model instead of GPU
+    place = fluid.CPUPlace()
+    exe = fluid.Executor(place)
+
+    model = create(main_arch)
+
+    startup_prog = fluid.Program()
+    infer_prog = fluid.Program()
+    with fluid.program_guard(infer_prog, startup_prog):
+        with fluid.unique_name.guard():
+            _, feed_vars = create_feed(test_feed, use_pyreader=False)
+            test_fetches = model.test(feed_vars)
+    infer_prog = infer_prog.clone(True)
+
+    exe.run(startup_prog)
+    checkpoint.load_params(exe, infer_prog, cfg.weights)
+
+    save_infer_model(FLAGS, exe, feed_vars, test_fetches, infer_prog)
+
+
+if __name__ == '__main__':
+    parser = ArgsParser()
+    parser.add_argument(
+        "--output_dir",
+        type=str,
+        default="output",
+        help="Directory for storing the output model files.")
+    FLAGS = parser.parse_args()
+    main()

tools/infer.py

@@ -96,48 +96,6 @@ def get_test_images(infer_dir, infer_img):
     return images
 
 
-def prune_feed_vars(feeded_var_names, target_vars, prog):
-    """
-    Filter out feed variables which are not in program,
-    pruned feed variables are only used in post processing
-    on model output, which are not used in program, such
-    as im_id to identify image order, im_shape to clip bbox
-    in image.
-    """
-    exist_var_names = []
-    prog = prog.clone()
-    prog = prog._prune(targets=target_vars)
-    global_block = prog.global_block()
-    for name in feeded_var_names:
-        try:
-            v = global_block.var(name)
-            exist_var_names.append(str(v.name))
-        except Exception:
-            logger.info('save_inference_model pruned unused feed '
-                        'variables {}'.format(name))
-            pass
-    return exist_var_names
-
-
-def save_infer_model(FLAGS, exe, feed_vars, test_fetches, infer_prog):
-    cfg_name = os.path.basename(FLAGS.config).split('.')[0]
-    save_dir = os.path.join(FLAGS.output_dir, cfg_name)
-    feeded_var_names = [var.name for var in feed_vars.values()]
-    target_vars = list(test_fetches.values())
-    feeded_var_names = prune_feed_vars(feeded_var_names, target_vars,
-                                       infer_prog)
-    logger.info("Save inference model to {}, input: {}, output: "
-                "{}...".format(save_dir, feeded_var_names,
-                               [str(var.name) for var in target_vars]))
-    fluid.io.save_inference_model(
-        save_dir,
-        feeded_var_names=feeded_var_names,
-        target_vars=target_vars,
-        executor=exe,
-        main_program=infer_prog,
-        params_filename="__params__")
-
-
 def main():
     cfg = load_config(FLAGS.config)
 
@@ -180,9 +138,6 @@ def main():
     if cfg.weights:
         checkpoint.load_params(exe, infer_prog, cfg.weights)
 
-    if FLAGS.save_inference_model:
-        save_infer_model(FLAGS, exe, feed_vars, test_fetches, infer_prog)
-
     # parse infer fetches
     assert cfg.metric in ['COCO', 'VOC', 'WIDERFACE'], \
             "unknown metric type {}".format(cfg.metric)
@@ -300,11 +255,6 @@ if __name__ == '__main__':
         type=float,
         default=0.5,
         help="Threshold to reserve the result for visualization.")
-    parser.add_argument(
-        "--save_inference_model",
-        action='store_true',
-        default=False,
-        help="Save inference model in output_dir if True.")
     parser.add_argument(
         "--use_tb",
         type=bool,