From a7a2e1bb4b0efa55ac2af91358e8c8d2d20076a7 Mon Sep 17 00:00:00 2001
From: AlexeyAB <alexeyab84@gmail.com>
Date: Fri, 30 Mar 2018 10:19:53 +0000
Subject: [PATCH] Example of anchors generation for Yolo v3
---
README.md | 80 +++++++++++++++++++++++++---------------
1 files changed, 50 insertions(+), 30 deletions(-)
diff --git a/README.md b/README.md
index 3e87eab..0830e12 100644
--- a/README.md
+++ b/README.md
@@ -15,11 +15,13 @@
10. [Using Yolo9000](#using-yolo9000)
11. [How to use Yolo as DLL](#how-to-use-yolo-as-dll)
-|  |  https://arxiv.org/abs/1612.08242 |
+
+
+|  |  https://pjreddie.com/media/files/papers/YOLOv3.pdf |
|---|---|
-|  |  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
@@ -186,7 +188,7 @@
## How to train (Pascal VOC Data):
-1. Download pre-trained weights for the convolutional layers (76 MB): http://pjreddie.com/media/files/darknet19_448.conv.23 and put to the directory `build\darknet\x64`
+1. Download pre-trained weights for the convolutional layers (154 MB): http://pjreddie.com/media/files/darknet53.conv.74 and put to the directory `build\darknet\x64`
2. Download The Pascal VOC Data and unpack it to directory `build\darknet\x64\data\voc` will be created dir `build\darknet\x64\data\voc\VOCdevkit\`:
* http://pjreddie.com/media/files/VOCtrainval_11-May-2012.tar
@@ -201,9 +203,13 @@
5. Run command: `type 2007_train.txt 2007_val.txt 2012_*.txt > train.txt`
-6. Set `batch=64` and `subdivisions=8` in the file `yolo-voc.2.0.cfg`: [link](https://github.com/AlexeyAB/darknet/blob/master/build/darknet/x64/yolo-voc.2.0.cfg#L2)
+6. Set `batch=64` and `subdivisions=8` in the file `yolov3-voc.cfg`: [link](https://github.com/AlexeyAB/darknet/blob/ee38c6e1513fb089b35be4ffa692afd9b3f65747/cfg/yolov3-voc.cfg#L3-L4)
-7. Start training by using `train_voc.cmd` or by using the command line: `darknet.exe detector train data/voc.data yolo-voc.2.0.cfg darknet19_448.conv.23` (**Note:** To disable Loss-Window use flag `-dont_show`. If you are using CPU, try `darknet_no_gpu.exe` instead of `darknet.exe`.)
+7. Start training by using `train_voc.cmd` or by using the command line:
+
+ `darknet.exe detector train data/voc.data cfg/yolov3-voc.cfg darknet53.conv.74`
+
+(**Note:** To disable Loss-Window use flag `-dont_show`. If you are using CPU, try `darknet_no_gpu.exe` instead of `darknet.exe`.)
If required change pathes in the file `build\darknet\x64\data\voc.data`
@@ -211,28 +217,38 @@
## How to train with multi-GPU:
-1. Train it first on 1 GPU for like 1000 iterations: `darknet.exe detector train data/voc.data yolo-voc.2.0.cfg darknet19_448.conv.23`
+1. Train it first on 1 GPU for like 1000 iterations: `darknet.exe detector train data/voc.data cfg/yolov3-voc.cfg darknet53.conv.74`
-2. Then stop and by using partially-trained model `/backup/yolo-voc_1000.weights` run training with multigpu (up to 4 GPUs): `darknet.exe detector train data/voc.data yolo-voc.2.0.cfg /backup/yolo-voc_1000.weights -gpus 0,1,2,3`
+2. Then stop and by using partially-trained model `/backup/yolo-voc_1000.weights` run training with multigpu (up to 4 GPUs): `darknet.exe detector train data/voc.data cfg/yolov3-voc.cfg /backup/yolo-voc_1000.weights -gpus 0,1,2,3`
https://groups.google.com/d/msg/darknet/NbJqonJBTSY/Te5PfIpuCAAJ
## How to train (to detect your custom objects):
-1. Create file `yolo-obj.cfg` with the same content as in `yolo-voc.2.0.cfg` (or copy `yolo-voc.2.0.cfg` to `yolo-obj.cfg)` and:
+1. Create file `yolo-obj.cfg` with the same content as in `yolov3.cfg` (or copy `yolov3.cfg` to `yolo-obj.cfg)` and:
- * change line batch to [`batch=64`](https://github.com/AlexeyAB/darknet/blob/master/build/darknet/x64/yolo-voc.2.0.cfg#L2)
- * change line subdivisions to [`subdivisions=8`](https://github.com/AlexeyAB/darknet/blob/master/build/darknet/x64/yolo-voc.2.0.cfg#L3)
- * change line `classes=20` to your number of objects
- * change line #237 from [`filters=125`](https://github.com/AlexeyAB/darknet/blob/master/cfg/yolo-voc.2.0.cfg#L224) to: filters=(classes + 5)x5, so if `classes=2` then should be `filters=35`. Or if you use `classes=1` then write `filters=30`, **do not write in the cfg-file: filters=(classes + 5)x5**.
+ * change line batch to [`batch=64`](https://github.com/AlexeyAB/darknet/blob/0039fd26786ab5f71d5af725fc18b3f521e7acfd/cfg/yolov3.cfg#L3)
+ * change line subdivisions to [`subdivisions=8`](https://github.com/AlexeyAB/darknet/blob/0039fd26786ab5f71d5af725fc18b3f521e7acfd/cfg/yolov3.cfg#L4)
+ * change line `classes=80` to your number of objects in each of 3 `[yolo]`-layers:
+ * https://github.com/AlexeyAB/darknet/blob/0039fd26786ab5f71d5af725fc18b3f521e7acfd/cfg/yolov3.cfg#L610
+ * https://github.com/AlexeyAB/darknet/blob/0039fd26786ab5f71d5af725fc18b3f521e7acfd/cfg/yolov3.cfg#L696
+ * https://github.com/AlexeyAB/darknet/blob/0039fd26786ab5f71d5af725fc18b3f521e7acfd/cfg/yolov3.cfg#L783
+ * change [`filters=255`] to filters=(classes + 5)x3 in the 3 `[convolutional]` before each `[yolo]` layer
+ * https://github.com/AlexeyAB/darknet/blob/0039fd26786ab5f71d5af725fc18b3f521e7acfd/cfg/yolov3.cfg#L603
+ * https://github.com/AlexeyAB/darknet/blob/0039fd26786ab5f71d5af725fc18b3f521e7acfd/cfg/yolov3.cfg#L689
+ * https://github.com/AlexeyAB/darknet/blob/0039fd26786ab5f71d5af725fc18b3f521e7acfd/cfg/yolov3.cfg#L776
+
+ So if `classes=1` then should be `filters=18`. If `classes=2` then write `filters=31`.
- (Generally `filters` depends on the `classes`, `num` and `coords`, i.e. equal to `(classes + coords + 1)*num`, where `num` is number of anchors)
+ **(Do not write in the cfg-file: filters=(classes + 5)x3)**
+
+ (Generally `filters` depends on the `classes`, `coords` and number of `mask`s, i.e. filters=`(classes + coords + 1)*<number of mask>`, where `mask` is indices of anchors. If `mask` is absence, then filters=`(classes + coords + 1)*num`)
- So for example, for 2 objects, your file `yolo-obj.cfg` should differ from `yolo-voc.2.0.cfg` in such lines:
+ So for example, for 2 objects, your file `yolo-obj.cfg` should differ from `yolov3.cfg` in such lines in each of **3** [yolo]-layers:
```
[convolutional]
- filters=35
+ filters=21
[region]
classes=2
@@ -276,12 +292,12 @@
data/obj/img3.jpg
```
-7. Download pre-trained weights for the convolutional layers (76 MB): http://pjreddie.com/media/files/darknet19_448.conv.23 and put to the directory `build\darknet\x64`
+7. Download pre-trained weights for the convolutional layers (154 MB): https://pjreddie.com/media/files/darknet53.conv.74 and put to the directory `build\darknet\x64`
-8. Start training by using the command line: `darknet.exe detector train data/obj.data yolo-obj.cfg darknet19_448.conv.23`
+8. Start training by using the command line: `darknet.exe detector train data/obj.data yolo-obj.cfg darknet53.conv.74`
(file `yolo-obj_xxx.weights` will be saved to the `build\darknet\x64\backup\` for each 100 iterations)
- (To disable Loss-Window use `darknet.exe detector train data/obj.data yolo-obj.cfg darknet19_448.conv.23 -dont_show`, if you train on computer without monitor like a cloud Amazaon EC2)
+ (To disable Loss-Window use `darknet.exe detector train data/obj.data yolo-obj.cfg darknet53.conv.74 -dont_show`, if you train on computer without monitor like a cloud Amazaon EC2)
9. After training is complete - get result `yolo-obj_final.weights` from path `build\darknet\x64\backup\`
@@ -304,7 +320,7 @@
Usually sufficient 2000 iterations for each class(object). But for a more precise definition when you should stop training, use the following manual:
-1. During training, you will see varying indicators of error, and you should stop when no longer decreases **0.060730 avg**:
+1. During training, you will see varying indicators of error, and you should stop when no longer decreases **0.XXXXXXX avg**:
> Region Avg IOU: 0.798363, Class: 0.893232, Obj: 0.700808, No Obj: 0.004567, Avg Recall: 1.000000, count: 8
> Region Avg IOU: 0.800677, Class: 0.892181, Obj: 0.701590, No Obj: 0.004574, Avg Recall: 1.000000, count: 8
@@ -355,7 +371,7 @@
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%
@@ -376,24 +392,28 @@
## How to improve object detection:
1. Before training:
- * set flag `random=1` in your `.cfg`-file - it will increase precision by training Yolo for different resolutions: [link]https://github.com/AlexeyAB/darknet/blob/master/cfg/yolo-voc.2.0.cfg#L244)
+ * set flag `random=1` in your `.cfg`-file - it will increase precision by training Yolo for different resolutions: [link](https://github.com/AlexeyAB/darknet/blob/0039fd26786ab5f71d5af725fc18b3f521e7acfd/cfg/yolov3.cfg#L788)
* increase network resolution in your `.cfg`-file (`height=608`, `width=608` or any value multiple of 32) - it will increase precision
+ * recalculate anchors for your dataset for `width` and `height` from cfg-file:
+ `darknet.exe detector calc_anchors data/obj.data -num_of_clusters 9 -width 416 -heigh 416`
+ then set the same 9 `anchors` in each of 3 `[yolo]`-layers in your cfg-file
+
* desirable that your training dataset include images with objects at diffrent: scales, rotations, lightings, from different sides
* desirable that your training dataset include images with objects (without labels) that you do not want to detect - negative samples
* for training with a large number of objects in each image, add the parameter `max=200` or higher value in the last layer [region] in your cfg-file
- * to speedup training (with decreasing detection accuracy) do Fine-Tuning instead of Transfer-Learning, set param `stopbackward=1` in one of the penultimate convolutional layers, for example here: https://github.com/AlexeyAB/darknet/blob/cad4d1618fee74471d335314cb77070fee951a42/cfg/yolo-voc.2.0.cfg#L202
+ * to speedup training (with decreasing detection accuracy) do Fine-Tuning instead of Transfer-Learning, set param `stopbackward=1` in one of the penultimate convolutional layers before the 1-st `[yolo]`-layer, for example here: https://github.com/AlexeyAB/darknet/blob/0039fd26786ab5f71d5af725fc18b3f521e7acfd/cfg/yolov3.cfg#L598
2. After training - for detection:
- * Increase network-resolution by set in your `.cfg`-file (`height=608` and `width=608`) or (`height=832` and `width=832`) or (any value multiple of 32) - this increases the precision and makes it possible to detect small objects: [link](https://github.com/AlexeyAB/darknet/blob/master/cfg/yolo-voc.2.0.cfg#L4)
+ * Increase network-resolution by set in your `.cfg`-file (`height=608` and `width=608`) or (`height=832` and `width=832`) or (any value multiple of 32) - this increases the precision and makes it possible to detect small objects: [link](https://github.com/AlexeyAB/darknet/blob/0039fd26786ab5f71d5af725fc18b3f521e7acfd/cfg/yolov3.cfg#L8-L9)
* you do not need to train the network again, just use `.weights`-file already trained for 416x416 resolution
- * if error `Out of memory` occurs then in `.cfg`-file you should increase `subdivisions=16`, 32 or 64: [link](https://github.com/AlexeyAB/darknet/blob/master/cfg/yolo-voc.2.0.cfg#L3)
+ * if error `Out of memory` occurs then in `.cfg`-file you should increase `subdivisions=16`, 32 or 64: [link](https://github.com/AlexeyAB/darknet/blob/0039fd26786ab5f71d5af725fc18b3f521e7acfd/cfg/yolov3.cfg#L4)
## How to mark bounded boxes of objects and create annotation files:
--
Gitblit v1.10.0