Merge branch 'master' of github.com:AlexeyAB/darknet

AlexeyAB

2018-03-28 6d56c38e8bcb9041335b03f27c192c24dfaedb1c

Merge branch 'master' of github.com:AlexeyAB/darknet

 README.md

@@ -15,11 +15,13 @@
10. [Using Yolo9000](#using-yolo9000)
11. [How to use Yolo as DLL](#how-to-use-yolo-as-dll)

|  ![Darknet Logo](http://pjreddie.com/media/files/darknet-black-small.png) |   ![map_fps](https://hsto.org/files/a24/21e/068/a2421e0689fb43f08584de9d44c2215f.jpg) https://arxiv.org/abs/1612.08242 |


|  ![Darknet Logo](http://pjreddie.com/media/files/darknet-black-small.png) |   ![map_fps](https://hsto.org/webt/pw/zd/0j/pwzd0jb9g7znt_dbsyw9qzbnvti.jpeg) https://pjreddie.com/media/files/papers/YOLOv3.pdf |
|---|---|

|  ![Darknet Logo](http://pjreddie.com/media/files/darknet-black-small.png) |   ![map_fps](https://hsto.org/files/3a6/fdf/b53/3a6fdfb533f34cee9b52bdd9bb0b19d9.jpg) https://arxiv.org/abs/1612.08242 |
|---|---|
* Yolo v2 on Pascal VOC 2007: https://hsto.org/files/a24/21e/068/a2421e0689fb43f08584de9d44c2215f.jpg
* Yolo v2 on Pascal VOC 2012 (comp4): https://hsto.org/files/3a6/fdf/b53/3a6fdfb533f34cee9b52bdd9bb0b19d9.jpg


# "You Only Look Once: Unified, Real-Time Object Detection (versions 2 & 3)"
@@ -46,11 +48,11 @@
* **GPU with CC >= 2.0** if you use CUDA, or **GPU CC >= 3.0** if you use cuDNN + CUDA: https://en.wikipedia.org/wiki/CUDA#GPUs_supported

##### Pre-trained models for different cfg-files can be downloaded from (smaller -> faster & lower quality):
* `yolov3.cfg` (236 MB COCO-model **v3**) - require 4 GB GPU-RAM: https://pjreddie.com/media/files/yolov3.weights
* `yolov2.cfg` (194 MB COCO-model v2) - require 4 GB GPU-RAM: https://pjreddie.com/media/files/yolov2.weights
* `yolo-voc.cfg` (194 MB VOC-model v2) - require 4 GB GPU-RAM: http://pjreddie.com/media/files/yolo-voc.weights
* `yolov2-tiny.cfg` (43 MB COCO-model v2) - require 1 GB GPU-RAM: https://pjreddie.com/media/files/yolov2-tiny.weights
* `yolov2-tiny-voc.cfg` (60 MB VOC-model v2) - require 1 GB GPU-RAM: http://pjreddie.com/media/files/yolov2-tiny-voc.weights
* `yolov3.cfg` (236 MB COCO **Yolo v3**) - require 4 GB GPU-RAM: https://pjreddie.com/media/files/yolov3.weights
* `yolov2.cfg` (194 MB COCO Yolo v2) - require 4 GB GPU-RAM: https://pjreddie.com/media/files/yolov2.weights
* `yolo-voc.cfg` (194 MB VOC Yolo v2) - require 4 GB GPU-RAM: http://pjreddie.com/media/files/yolo-voc.weights
* `yolov2-tiny.cfg` (43 MB COCO Yolo v2) - require 1 GB GPU-RAM: https://pjreddie.com/media/files/yolov2-tiny.weights
* `yolov2-tiny-voc.cfg` (60 MB VOC Yolo v2) - require 1 GB GPU-RAM: http://pjreddie.com/media/files/yolov2-tiny-voc.weights
* `yolo9000.cfg` (186 MB Yolo9000-model) - require 4 GB GPU-RAM: http://pjreddie.com/media/files/yolo9000.weights

Put it near compiled: darknet.exe
@@ -355,7 +357,7 @@

1. To calculate mAP (mean average precision) on PascalVOC-2007-test:
* Download PascalVOC dataset, install Python 3.x and get file `2007_test.txt` as described here: https://github.com/AlexeyAB/darknet#how-to-train-pascal-voc-data
* Then download file https://raw.githubusercontent.com/AlexeyAB/darknet/master/scripts/voc_label_difficult.py to the dir `build\darknet\x64\data\voc` then run `voc_label_difficult.py` to get the file `difficult_2007_test.txt`
* Then download file https://raw.githubusercontent.com/AlexeyAB/darknet/master/scripts/voc_label_difficult.py to the dir `build\darknet\x64\data\` then run `voc_label_difficult.py` to get the file `difficult_2007_test.txt`
* Remove symbol `#` from this line to un-comment it: https://github.com/AlexeyAB/darknet/blob/master/build/darknet/x64/data/voc.data#L4
* Then there are 2 ways to get mAP:
    1. Using Darknet + Python: run the file `build/darknet/x64/calc_mAP_voc_py.cmd` - you will get mAP for `yolo-voc.cfg` model, mAP = 75.9%