:fire: :fire: yolo v2 :fire: :fire:

Joseph Redmon

2016-11-17 c6afc7ff1499fbbe64069e1843d7929bd7ae2eaa

:fire: :fire: yolo v2 :fire: :fire:

 cfg/coco.data

New file
@@ -0,0 +1,8 @@
classes= 80
train  = /home/pjreddie/data/coco/trainvalno5k.txt
valid  = coco_testdev
#valid = data/coco_val_5k.list
names = data/coco.names
backup = /home/pjreddie/backup/
eval=coco


 cfg/imagenet1k.data

 cfg/voc.data

New file
@@ -0,0 +1,6 @@
classes= 20
train  = /home/pjreddie/data/voc/train.txt
valid  = /home/pjreddie/data/voc/2007_test.txt
names = data/pascal.names
backup = /home/pjreddie/backup/


 cfg/yolo.cfg

@@ -1,36 +1,25 @@
[net]
batch=1
subdivisions=1
height=448
width=448
batch=64
subdivisions=8
height=416
width=416
channels=3
momentum=0.9
decay=0.0005
saturation=1.5
exposure=1.5
angle=0
saturation = 1.5
exposure = 1.5
hue=.1

learning_rate=0.0005
learning_rate=0.001
max_batches = 120000
policy=steps
steps=200,400,600,20000,30000
scales=2.5,2,2,.1,.1
max_batches = 40000
steps=-1,100,80000,100000
scales=.1,10,.1,.1

[convolutional]
batch_normalize=1
filters=64
size=7
stride=2
pad=1
activation=leaky

[maxpool]
size=2
stride=2

[convolutional]
batch_normalize=1
filters=192
filters=32
size=3
stride=1
pad=1
@@ -42,6 +31,54 @@

[convolutional]
batch_normalize=1
filters=64
size=3
stride=1
pad=1
activation=leaky

[maxpool]
size=2
stride=2

[convolutional]
batch_normalize=1
filters=128
size=3
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=64
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=128
size=3
stride=1
pad=1
activation=leaky

[maxpool]
size=2
stride=2

[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
@@ -56,6 +93,34 @@
pad=1
activation=leaky

[maxpool]
size=2
stride=2

[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=256
@@ -78,88 +143,12 @@

[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=512
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=1024
size=3
stride=1
pad=1
activation=leaky

[maxpool]
size=2
stride=2

[convolutional]
batch_normalize=1
filters=512
@@ -192,6 +181,7 @@
pad=1
activation=leaky


#######

[convolutional]
@@ -205,10 +195,19 @@
[convolutional]
batch_normalize=1
size=3
stride=1
pad=1
filters=1024
activation=leaky

[route]
layers=-9

[reorg]
stride=2
pad=1
filters=1024
activation=leaky

[route]
layers=-1,-3

[convolutional]
batch_normalize=1
@@ -219,39 +218,27 @@
activation=leaky

[convolutional]
batch_normalize=1
size=3
size=1
stride=1
pad=1
filters=1024
activation=leaky

[local]
size=3
stride=1
pad=1
filters=256
activation=leaky

[dropout]
probability=.5

[connected]
output= 1715
filters=425
activation=linear

[detection]
classes=20
[region]
anchors = 0.738768,0.874946,  2.42204,2.65704,  4.30971,7.04493,  10.246,4.59428,  12.6868,11.8741
bias_match=1
classes=80
coords=4
rescore=1
side=7
num=3
softmax=0
sqrt=1
num=5
softmax=1
jitter=.2
rescore=1

object_scale=1
noobject_scale=.5
object_scale=5
noobject_scale=1
class_scale=1
coord_scale=5
coord_scale=1

absolute=1
thresh = .6
random=0

 cfg/yolo_voc.cfg

New file
@@ -0,0 +1,244 @@
[net]
batch=64
subdivisions=8
height=416
width=416
channels=3
momentum=0.9
decay=0.0005
angle=0
saturation = 1.5
exposure = 1.5
hue=.1

learning_rate=0.0001
max_batches = 45000
policy=steps
steps=100,25000,35000
scales=10,.1,.1

[convolutional]
batch_normalize=1
filters=32
size=3
stride=1
pad=1
activation=leaky

[maxpool]
size=2
stride=2

[convolutional]
batch_normalize=1
filters=64
size=3
stride=1
pad=1
activation=leaky

[maxpool]
size=2
stride=2

[convolutional]
batch_normalize=1
filters=128
size=3
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=64
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=128
size=3
stride=1
pad=1
activation=leaky

[maxpool]
size=2
stride=2

[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
pad=1
activation=leaky

[maxpool]
size=2
stride=2

[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky

[maxpool]
size=2
stride=2

[convolutional]
batch_normalize=1
filters=1024
size=3
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=512
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=1024
size=3
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=512
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=1024
size=3
stride=1
pad=1
activation=leaky


#######

[convolutional]
batch_normalize=1
size=3
stride=1
pad=1
filters=1024
activation=leaky

[convolutional]
batch_normalize=1
size=3
stride=1
pad=1
filters=1024
activation=leaky

[route]
layers=-9

[reorg]
stride=2

[route]
layers=-1,-3

[convolutional]
batch_normalize=1
size=3
stride=1
pad=1
filters=1024
activation=leaky

[convolutional]
size=1
stride=1
pad=1
filters=125
activation=linear

[region]
anchors = 1.08,1.19,  3.42,4.41,  6.63,11.38,  9.42,5.11,  16.62,10.52
bias_match=1
classes=20
coords=4
num=5
softmax=1
jitter=.2
rescore=1

object_scale=5
noobject_scale=1
class_scale=1
coord_scale=1

absolute=1
thresh = .6
random=0

 cfg/yolov1/tiny-coco.cfg

 cfg/yolov1/tiny-yolo.cfg

 cfg/yolov1/xyolo.test.cfg

 cfg/yolov1/yolo-coco.cfg

 cfg/yolov1/yolo-small.cfg

 cfg/yolov1/yolo.cfg

New file
@@ -0,0 +1,257 @@
[net]
batch=1
subdivisions=1
height=448
width=448
channels=3
momentum=0.9
decay=0.0005
saturation=1.5
exposure=1.5
hue=.1

learning_rate=0.0005
policy=steps
steps=200,400,600,20000,30000
scales=2.5,2,2,.1,.1
max_batches = 40000

[convolutional]
batch_normalize=1
filters=64
size=7
stride=2
pad=1
activation=leaky

[maxpool]
size=2
stride=2

[convolutional]
batch_normalize=1
filters=192
size=3
stride=1
pad=1
activation=leaky

[maxpool]
size=2
stride=2

[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky

[maxpool]
size=2
stride=2

[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=512
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=1024
size=3
stride=1
pad=1
activation=leaky

[maxpool]
size=2
stride=2

[convolutional]
batch_normalize=1
filters=512
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=1024
size=3
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=512
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=1024
size=3
stride=1
pad=1
activation=leaky

#######

[convolutional]
batch_normalize=1
size=3
stride=1
pad=1
filters=1024
activation=leaky

[convolutional]
batch_normalize=1
size=3
stride=2
pad=1
filters=1024
activation=leaky

[convolutional]
batch_normalize=1
size=3
stride=1
pad=1
filters=1024
activation=leaky

[convolutional]
batch_normalize=1
size=3
stride=1
pad=1
filters=1024
activation=leaky

[local]
size=3
stride=1
pad=1
filters=256
activation=leaky

[dropout]
probability=.5

[connected]
output= 1715
activation=linear

[detection]
classes=20
coords=4
rescore=1
side=7
num=3
softmax=0
sqrt=1
jitter=.2

object_scale=1
noobject_scale=.5
class_scale=1
coord_scale=5


 cfg/yolov1/yolo.train.cfg

 cfg/yolov1/yolo2.cfg

 data/coco.names

New file
@@ -0,0 +1,80 @@
person
bicycle
car
motorbike
aeroplane
bus
train
truck
boat
traffic light
fire hydrant
stop sign
parking meter
bench
bird
cat
dog
horse
sheep
cow
elephant
bear
zebra
giraffe
backpack
umbrella
handbag
tie
suitcase
frisbee
skis
snowboard
sports ball
kite
baseball bat
baseball glove
skateboard
surfboard
tennis racket
bottle
wine glass
cup
fork
knife
spoon
bowl
banana
apple
sandwich
orange
broccoli
carrot
hot dog
pizza
donut
cake
chair
sofa
pottedplant
bed
diningtable
toilet
tvmonitor
laptop
mouse
remote
keyboard
cell phone
microwave
oven
toaster
sink
refrigerator
book
clock
vase
scissors
teddy bear
hair drier
toothbrush

 src/blas.c

@@ -5,6 +5,28 @@
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
void reorg_cpu(float *x, int w, int h, int c, int batch, int stride, int forward, float *out)
{
    int b,i,j,k;
    int out_c = c/(stride*stride);

    for(b = 0; b < batch; ++b){
        for(k = 0; k < c; ++k){
            for(j = 0; j < h; ++j){
                for(i = 0; i < w; ++i){
                    int in_index  = i + w*(j + h*(k + c*b));
                    int c2 = k % out_c;
                    int offset = k / out_c;
                    int w2 = i*stride + offset % stride;
                    int h2 = j*stride + offset / stride;
                    int out_index = w2 + w*stride*(h2 + h*stride*(c2 + out_c*b));
                    if(forward) out[out_index] = x[in_index];
                    else out[in_index] = x[out_index];
                }
            }
        }
    }
}

void flatten(float *x, int size, int layers, int batch, int forward)
{

 src/blas.h

@@ -4,6 +4,7 @@
void pm(int M, int N, float *A);
float *random_matrix(int rows, int cols);
void time_random_matrix(int TA, int TB, int m, int k, int n);
void reorg_cpu(float *x, int w, int h, int c, int batch, int stride, int forward, float *out);

void test_blas();


 src/darknet.c

@@ -13,6 +13,7 @@
#endif

extern void predict_classifier(char *datacfg, char *cfgfile, char *weightfile, char *filename, int top);
extern void test_detector(char *datacfg, char *cfgfile, char *weightfile, char *filename, float thresh);
extern void run_voxel(int argc, char **argv);
extern void run_yolo(int argc, char **argv);
extern void run_detector(int argc, char **argv);
@@ -379,6 +380,10 @@
        run_super(argc, argv);
    } else if (0 == strcmp(argv[1], "detector")){
        run_detector(argc, argv);
    } else if (0 == strcmp(argv[1], "detect")){
        float thresh = find_float_arg(argc, argv, "-thresh", .25);
        char *filename = (argc > 4) ? argv[4]: 0;
        test_detector("cfg/coco.data", argv[2], argv[3], filename, thresh);
    } else if (0 == strcmp(argv[1], "cifar")){
        run_cifar(argc, argv);
    } else if (0 == strcmp(argv[1], "go")){
@@ -390,7 +395,7 @@
    } else if (0 == strcmp(argv[1], "coco")){
        run_coco(argc, argv);
    } else if (0 == strcmp(argv[1], "classify")){
        predict_classifier("cfg/imagenet1k.dataset", argv[2], argv[3], argv[4], 5);
        predict_classifier("cfg/imagenet1k.data", argv[2], argv[3], argv[4], 5);
    } else if (0 == strcmp(argv[1], "classifier")){
        run_classifier(argc, argv);
    } else if (0 == strcmp(argv[1], "art")){

 src/demo.c

@@ -110,6 +110,7 @@
    srand(2222222);

    if(filename){
        printf("video file: %s\n", filename);
        cap = cvCaptureFromFile(filename);
    }else{
        cap = cvCaptureFromCAM(cam_index);

 src/detector.c

@@ -490,7 +490,7 @@
void run_detector(int argc, char **argv)
{
    char *prefix = find_char_arg(argc, argv, "-prefix", 0);
    float thresh = find_float_arg(argc, argv, "-thresh", .2);
    float thresh = find_float_arg(argc, argv, "-thresh", .25);
    int cam_index = find_int_arg(argc, argv, "-c", 0);
    int frame_skip = find_int_arg(argc, argv, "-s", 0);
    if(argc < 4){

 src/image.c

@@ -185,10 +185,16 @@
        int class = max_index(probs[i], classes);
        float prob = probs[i][class];
        if(prob > thresh){
            //int width = pow(prob, 1./2.)*30+1;

            int width = im.h * .012;

            if(0){
                width = pow(prob, 1./2.)*10+1;
                alphabet = 0;
            }

            printf("%s: %.0f%%\n", names[class], prob*100);
            int offset = class*1 % classes;
            int offset = class*123457 % classes;
            float red = get_color(2,offset,classes);
            float green = get_color(1,offset,classes);
            float blue = get_color(0,offset,classes);

 src/parser.c

@@ -238,9 +238,6 @@
    int classes = option_find_int(options, "classes", 20);
    int num = option_find_int(options, "num", 1);

    params.w = option_find_int(options, "side", params.w);
    params.h = option_find_int(options, "side", params.h);

    layer l = make_region_layer(params.batch, params.w, params.h, num, classes, coords);
    assert(l.outputs == params.inputs);


 src/region_layer.c

@@ -44,7 +44,7 @@
    l.delta_gpu = cuda_make_array(l.delta, batch*l.outputs);
#endif

    fprintf(stderr, "Region Layer\n");
    fprintf(stderr, "detection\n");
    srand(0);

    return l;

 src/reorg_layer.c

@@ -23,7 +23,7 @@
        l.out_c = c*(stride*stride);
    }
    l.reverse = reverse;
    fprintf(stderr, "Reorg Layer: %d x %d x %d image -> %d x %d x %d image, \n", w,h,c,l.out_w, l.out_h, l.out_c);
    fprintf(stderr, "reorg              /%2d  %4d x%4d x%4d   ->  %4d x%4d x%4d\n",  stride, w, h, c, l.out_w, l.out_h, l.out_c);
    l.outputs = l.out_h * l.out_w * l.out_c;
    l.inputs = h*w*c;
    int output_size = l.out_h * l.out_w * l.out_c * batch;
@@ -77,45 +77,19 @@

void forward_reorg_layer(const layer l, network_state state)
{
    int b,i,j,k;

    for(b = 0; b < l.batch; ++b){
        for(k = 0; k < l.c; ++k){
            for(j = 0; j < l.h; ++j){
                for(i = 0; i < l.w; ++i){
                    int in_index  = i + l.w*(j + l.h*(k + l.c*b));

                    int c2 = k % l.out_c;
                    int offset = k / l.out_c;
                    int w2 = i*l.stride + offset % l.stride;
                    int h2 = j*l.stride + offset / l.stride;
                    int out_index = w2 + l.out_w*(h2 + l.out_h*(c2 + l.out_c*b));
                    l.output[out_index] = state.input[in_index];
                }
            }
        }
    if(l.reverse){
        reorg_cpu(state.input, l.w, l.h, l.c, l.batch, l.stride, 1, l.output);
    }else {
        reorg_cpu(state.input, l.w, l.h, l.c, l.batch, l.stride, 0, l.output);
    }
}

void backward_reorg_layer(const layer l, network_state state)
{
    int b,i,j,k;

    for(b = 0; b < l.batch; ++b){
        for(k = 0; k < l.c; ++k){
            for(j = 0; j < l.h; ++j){
                for(i = 0; i < l.w; ++i){
                    int in_index  = i + l.w*(j + l.h*(k + l.c*b));

                    int c2 = k % l.out_c;
                    int offset = k / l.out_c;
                    int w2 = i*l.stride + offset % l.stride;
                    int h2 = j*l.stride + offset / l.stride;
                    int out_index = w2 + l.out_w*(h2 + l.out_h*(c2 + l.out_c*b));
                    state.delta[in_index] = l.delta[out_index];
                }
            }
        }
    if(l.reverse){
        reorg_cpu(l.delta, l.w, l.h, l.c, l.batch, l.stride, 0, state.delta);
    }else{
        reorg_cpu(l.delta, l.w, l.h, l.c, l.batch, l.stride, 1, state.delta);
    }
}


 src/route_layer.c

@@ -5,7 +5,7 @@

route_layer make_route_layer(int batch, int n, int *input_layers, int *input_sizes)
{
    fprintf(stderr,"Route Layer:");
    fprintf(stderr,"route ");
    route_layer l = {0};
    l.type = ROUTE;
    l.batch = batch;