COCO dataset for SSD and update README.md #844

sefira · 2018-04-13T12:25:41Z

No description provided.

…to ssd_coco_reader

… ssd_coco_reader

…or_test=True)

qingqing01 · 2018-04-20T09:38:18Z

fluid/object_detection/eval.py

@@ -13,27 +13,27 @@
 parser = argparse.ArgumentParser(description=__doc__)
 add_arg = functools.partial(add_arguments, argparser=parser)
 # yapf: disable
-add_arg('dataset',          str, 'pascalvoc', "coco or pascalvoc.")
+add_arg('dataset',          str,   'coco2014',  "coco2014, coco2017, and pascalvoc.")


We can use pascalvoc as default training for all arguments.

qingqing01 · 2018-04-20T09:38:48Z

fluid/object_detection/eval.py

-add_arg('mean_value_R',     float, 127.5,      "mean value for R channel which will be subtracted")  #103.94
+add_arg('model_dir',        str,   '',     "The model path.")
+add_arg('nms_threshold',    float, 0.5,    "nms threshold")
+add_arg('ap_version',       str,   'integral',   "integral, 11points, and cocoMAP")


The VOC use 11point by default.

qingqing01 · 2018-04-20T09:39:08Z

fluid/object_detection/eval.py

+add_arg('nms_threshold',    float, 0.5,    "nms threshold")
+add_arg('ap_version',       str,   'integral',   "integral, 11points, and cocoMAP")
+add_arg('resize_h',         int,   300,    "resize image size")
+add_arg('resize_w',         int,   300,    "resize image size")


resize image size -> The resized image width.

qingqing01 · 2018-04-20T09:39:21Z

fluid/object_detection/eval.py

+add_arg('model_dir',        str,   '',     "The model path.")
+add_arg('nms_threshold',    float, 0.5,    "nms threshold")
+add_arg('ap_version',       str,   'integral',   "integral, 11points, and cocoMAP")
+add_arg('resize_h',         int,   300,    "resize image size")


resize image size -> The resized image height.

qingqing01 · 2018-04-20T09:39:47Z

fluid/object_detection/eval.py

+add_arg('resize_w',         int,   300,    "resize image size")
+add_arg('mean_value_B',     float, 127.5,  "mean value for B channel which will be subtracted")  #123.68
+add_arg('mean_value_G',     float, 127.5,  "mean value for G channel which will be subtracted")  #116.78
+add_arg('mean_value_R',     float, 127.5,  "mean value for R channel which will be subtracted")  #103.94
 # yapf: enable


yapf: enable -> yapf: enable

qingqing01 · 2018-04-20T10:42:50Z

fluid/object_detection/reader.py

-    else:
-        return pascalvoc(settings, file_list, 'test', False)
+        return _reader_creator(test_settings, file_list, 'test', False)
+    elif 'pascalvoc' in settings.dataset:


qingqing01 · 2018-04-20T10:43:17Z

fluid/object_detection/train.py

+add_arg('mean_value_G',     float, 127.5, "mean value which will be subtracted")  #116.78
+add_arg('mean_value_R',     float, 127.5, "mean value which will be subtracted")  #103.94
+add_arg('is_toy',           int,   0, "Toy for quick debug, 0 means using all data, while n means using only n sample")
+# yapf: disable


yapf: disable -> yapf: enable

qingqing01 · 2018-04-20T10:43:49Z

fluid/object_detection/train.py

+add_arg('mean_value_B',     float, 127.5, "mean value which will be subtracted")  #123.68
+add_arg('mean_value_G',     float, 127.5, "mean value which will be subtracted")  #116.78
+add_arg('mean_value_R',     float, 127.5, "mean value which will be subtracted")  #103.94
+add_arg('is_toy',           int,   0, "Toy for quick debug, 0 means using all data, while n means using only n sample")


上面默认参数请统一按照VOC来设置。

qingqing01 · 2018-04-20T10:44:19Z

fluid/object_detection/train.py

+add_arg('nms_threshold',    float, 0.5,    "nms threshold")
+add_arg('ap_version',       str,   'integral',   "integral, 11points")
+add_arg('resize_h',         int,   300,    "resize image size")
+add_arg('resize_w',         int,   300,    "resize image size")


resize image size 语法有误，请按照左边的改过来。

qingqing01 · 2018-04-20T10:45:23Z

fluid/object_detection/reader.py

-                 apply_expand=True,
-                 toy=0):
+    def __init__(self, dataset, ap_version, toy, data_dir, label_file, resize_h,
+                 resize_w, mean_value, apply_distort, apply_expand):


参数顺序请按照左边的吧，一些可以设置默认值的。

qingqing01 · 2018-04-24T05:01:06Z

fluid/object_detection/README.md

  ```

 TBD

 ### Evaluate

 ```python
-env CUDA_VISIABLE_DEVICES=0 python eval.py --model='model/90' --test_list=''
+env CUDA_VISIABLE_DEVICES=0 python eval.py --dataset='pascalvoc' --model_dir='train_pascal_model/90' --test_list='' --ap_version='11point'
+env CUDA_VISIABLE_DEVICES=0 python eval_cocoMAP.py --dataset='coco2014' --model_dir='train_coco_model/24'
 ```


env CUDA_VISIABLE_DEVICES=0 python eval.py --model='model/90' --data_dir='data/pascalvoc' --test_list='test.txt'

VOC和COCO分开写吧。

coco的评估命令。

qingqing01 · 2018-04-24T05:03:14Z

fluid/object_detection/README.md

 ```
+You can evaluate your trained model in difference metric like 11point, integral on both PASCAL VOC and COCO dataset. Moreover, we provide eval_cocoMAP.py which uses a COCO-specific mAP metric defined by [COCO committee](http://cocodataset.org/#detections-eval). To use this eval_cocoMAP.py, [cocoapi](https://github.com/cocodataset/cocoapi) is needed.


in difference metric -> in different metric

eval_cocoMAP.py -> eval_coco.py 脚本名字也换下吧，觉得不要大小写混合。

告诉用户如何安装 cocoapi ？

qingqing01 · 2018-04-24T05:03:44Z

fluid/object_detection/README.md

+<img src="images/COCO_val2014_000000142324.jpg" height=300 width=400 hspace='10'/>
+<img src="images/COCO_val2014_000000144003.jpg" height=300 width=400 hspace='10'/> <br />
+MobileNet-SSD300x300 Visualization Examples
+</p>


这里是否要提供VOC的数据可视化？

qingqing01 · 2018-04-24T05:04:33Z

fluid/object_detection/eval_cocoMAP.py

@@ -0,0 +1,154 @@
+import os


eval_cocoMAP.py -> eval_coco_map.py ?

qingqing01 · 2018-04-24T05:05:17Z

fluid/object_detection/eval_cocoMAP.py

+        def if_exist(var):
+            return os.path.exists(os.path.join(model_dir, var.name))
+
+        fluid.io.load_vars(exe, model_dir, predicate=if_exist)


去掉 line 62和line65多余的空格。

That was added by yapf, revised as comments

qingqing01 · 2018-04-24T05:06:11Z

fluid/object_detection/eval_cocoMAP.py

+                        'bbox': bbox,
+                        'score': score
+                    }
+                    dts_res.append(dt_res)


line 85到line113数据重组是否要单独放个函数调用？使得代码结构清晰点。

qingqing01 · 2018-04-24T05:07:51Z

fluid/object_detection/infer.py

+        def if_exist(var):
+            return os.path.exists(os.path.join(model_dir, var.name))
+
+        fluid.io.load_vars(exe, model_dir, predicate=if_exist)


去掉多余的空格。

qingqing01 · 2018-04-24T05:08:45Z

fluid/object_detection/train.py

-add_arg('is_toy',           int, 0, "Toy for quick debug, 0 means using all data, while n means using only n sample")
-# yapf: enable
-
+add_arg('num_passes',       int,   25,        "Epoch number.")


25 -> 120 设置成VOC默认的吧。

qingqing01 · 2018-04-24T05:10:12Z

fluid/object_detection/train.py

        if test_map[0] > best_map:
            best_map = test_map[0]
            save_model('best_model')
-        print("Test {0}, map {1}".format(pass_id, test_map[0]))
+        print("Pass {0}, map {1}".format(pass_id, test_map[0]))


map -> test map ? log中体现是test集。

sefira added 30 commits March 21, 2018 19:51

ready to coco_reader

fc88336

complete coco_reader.py & coco_train.py

cd568b6

complete coco reader

e037b8e

Merge branch 'develop' into ssd_coco_reader

5336461

rename file

1e2b631

Merge branch 'develop' into ssd_coco_reader

997f042

use argparse instead of explicit assignment

1396550

fix

33faeb7

fix reader bug for some gray image in coco data

73a037f

ready to train coco

c164349

fix bug in test()

f4d7765

fix bug in test()

9fd6d88

change coco dataset to coco2017 dataset

dde4fcd

change dataset from coco to coco2017

c4e9e17

change learning rate

69afb99

fix bug in gt label (category id 2 label)

ed70cbf

fix bug in background label

027df47

save model when train finished

2cf5084

use coco map

4a059e7

Merge branch 'develop' into ssd_coco_reader

b910bf2

wqMerge branch 'develop' of https://github.com/PaddlePaddle/models in…

9d3f38e

…to ssd_coco_reader

adding coco year version args: 2014 or 2017

8c64c9c

Merge branch 'develop' of https://github.com/PaddlePaddle/models into…

5f6b2a9

… ssd_coco_reader

add coco dataset download, and README.md

61adbdb

fix

b016bc1

fix image truncted IOError, map version error

9e31023

add test config

26063fd

add eval.py for evaluate trained model

863415d

fix

58cae07

fix bug when cocoMAP

7ebd4bd

sefira and others added 9 commits April 19, 2018 16:43

updata READEME.md

6f4d939

fix cocoMAP bug

6bd3f3f

find strange with test_program = fluid.default_main_program().clone(f…

813fb39

…or_test=True)

add inference and visualize, awa, README.md

2d0d2ae

upload infer&visual example image

4efa01c

refine image

a09b41c

refine

932b150

Merge branch 'develop' into ssd_coco_reader

46db499

fix bug after merge

161aea8

sefira force-pushed the ssd_coco_reader branch from f927827 to 46db499 Compare April 20, 2018 03:08

follow yapf

b0bfbc3

qingqing01 reviewed Apr 20, 2018

View reviewed changes

qingqing01 changed the title ~~ssd coco dataset and readme~~ COCO dataset for SSD and update README.md Apr 20, 2018

sefira and others added 3 commits April 20, 2018 23:17

follow comments

8e0d855

fix bug after separate eval and eval_cocoMAP

88bc588

follow yapf

a08b8e5

qingqing01 reviewed Apr 24, 2018

View reviewed changes

sefira added 3 commits April 26, 2018 00:09

follow comments

a5e427f

follow yapf

ba0a42f

follow yapf

b016ed6

sefira force-pushed the ssd_coco_reader branch from 81b6707 to b016ed6 Compare April 25, 2018 16:37

qingqing01 approved these changes Apr 27, 2018

View reviewed changes

qingqing01 merged commit 496ff37 into PaddlePaddle:develop Apr 27, 2018

sefira deleted the ssd_coco_reader branch April 28, 2018 09:16

sefira restored the ssd_coco_reader branch April 28, 2018 09:17

		```
		You can evaluate your trained model in difference metric like 11point, integral on both PASCAL VOC and COCO dataset. Moreover, we provide eval_cocoMAP.py which uses a COCO-specific mAP metric defined by [COCO committee](http://cocodataset.org/#detections-eval). To use this eval_cocoMAP.py, [cocoapi](https://github.com/cocodataset/cocoapi) is needed.

COCO dataset for SSD and update README.md #844

COCO dataset for SSD and update README.md #844

Conversation

sefira commented Apr 13, 2018

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

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