Darknet: STB Reason: can't fopen/STB Reason: unknown image type Cannot load image - tried other solution

[sbanerj2]

I am getting the following error for some images (custom dataset) as a result of which the training stops.
(I am using /path/to/ as an abbreviation of my absolute path. All the files and images exist.)

Loading weights from ./darknet53.conv.74...Done!
Cannot load image "/path/to/Object_Detection/Ground/JPEGImages/Training/6/um_50_na_sunny_sony_0_384.jpg"
STB Reason: can't fopen
Cannot load image "/path/to/Object_Detection/Ground/JPEGImages/Training/6/um_50_na_sunny_sony_0_41.jpg"
STB Reason: unknown image type
Cannot load image "/path/to/Object_Detection/Ground/JPEGImages/Training/3/golf_40_na_cloudy_sony_0_273.jpg"
STB Reason: unknown image type

My images are present in the location too and their corresponding text files are in the following location:
/path/to/Object_Detection/Ground/labels/Training/...

The format of the folders are like Pascal VOC

here's a part of the makefile:
GPU=1
CUDNN=0
OPENCV=1
OPENMP=0
DEBUG=0

And here's my ground.data:

classes= 19
train  = /path/to/Object_Detection/ground_train.txt
valid  = /path/to/Object_Detection/ground_val.txt
names = /path/to/darknet/data/ground.names
backup = /path/to/darknet/backup/ground/

Here's the command I am using:

./darknet detector train /path/to/darknet/cfg/ground.data /path/to/darknet/cfg/yolov3-ground.cfg ./darknet53.conv.74 > /path/to/darknet/scripts/ground-train.log

Here's how my ground_train.txt (has all absolute path for images. only a portion of the entire file is shown) looks :

/path/to/Object_Detection/Ground/JPEGImages/Training/0/analog_50_180_sunny_sony_0_0.jpg
/path/to/Object_Detection/Ground/JPEGImages/Training/0/analog_50_180_sunny_sony_0_1.jpg
/path/to/Object_Detection/Ground/JPEGImages/Training/0/analog_50_180_sunny_sony_0_2.jpg
...

All the files exists and are valid. I am not sure what's wrong. It works for other files in the training dataset but only stops for these. I checked whether the images are of correct format, and they seem ok.

Here's the output after 14 epochs, my training stops after this:

14: 573.572327, 927.539124 avg, 0.000000 rate, 1636.650012 seconds, 896 images
Loaded: 0.000064 seconds
: 0.219096, Class: 0.386432, Obj: 0.530981, No Obj: 0.511523, .5R: 0.000000, .75R: 0.000000,  count: 2
Region 106 Avg IOU: 0.011220, Class: 0.282754, Obj: 0.311355, No Obj: 0.468680, .5R: 0.000000, .75R: 0.000000,  count: 1
Region 82 Avg IOU: 0.279148, Class: 0.148486, Obj: 0.629275, No Obj: 0.458073, .5R: 0.000000, .75R: 0.000000,  count: 1
Region 94 Avg IOU: 0.075961, Class: 0.271695, Obj: 0.082233, No Obj: 0.514065, .5R: 0.000000, .75R: 0.000000,  count: 2
Region 106 Avg IOU: 0.352088, Class: 0.543684, Obj: 0.793743, No Obj: 0.471246, .5R: 0.000000, .75R: 0.000000,  count: 1
Region 82 Avg IOU: 0.165622, Class: 0.815933, Obj: 0.489756, No Obj: 0.457557, .5R: 0.000000, .75R: 0.000000,  count: 1
Region 94 Avg IOU: 0.053709, Class: 0.664887, Obj: 0.541732, No Obj: 0.512764, .5R: 0.000000, .75R: 0.000000,  count: 2
Region 106 Avg IOU: 0.073817, Class: 0.247052, Obj: 0.413561, No Obj: 0.469873, .5R: 0.000000, .75R: 0.000000,  count: 1
14: 573.572327, 927.539124 avg, 0.000000 rate, 1636.650012 seconds, 896 images
Loaded: 0.000064 seconds

[Mps24-7uk]

I am getting the same issue Cannot load image "./darknet/dataset/z1.jpg" STB Reason: can't fopen @sbanerj2

[SalahAdDin]

I'm also getting the same problem.

[PhilipMachineLearning]

I'm getting the same issue on Windows using the bash shell.

[tijnonlijn]

The same happens here on ubuntu 18.04. Now here's the funny part: after retrying 5-10 times while randomly changing the file name it suddenly works. Even when I use files that previously gave an error. And once it works it remains working without any error.

Now I'm not really a linux expert, but could it be that there is some sort of background process that takes some time to start?

[sbanerj2]

For me the problem was compilation. I was compiling yolo with a wrong version of CUDA. Use the fork by alexeyab and make sure you have all the requirements. Make sure to update the makefile to

ifeq ($(GPU), 1)

COMMON+= -DGPU -I/usr/local/cuda/include/ -I/path/to/cudnn/7.0.5/cuda/include
CFLAGS+= -DGPU
ifeq ($(OS),Darwin) #MAC
LDFLAGS+= -L/usr/local/cuda/lib -lcuda -lcudart -lcublas -lcurand
else
LDFLAGS+= -L/path/to/cuda/ -lcuda -lcudart -lcublas -lcurand
endif
endif

ifeq ($(CUDNN), 1)
COMMON+= -DCUDNN
ifeq ($(OS),Darwin) #MAC
CFLAGS+= -DCUDNN -I/usr/local/cuda/include
LDFLAGS+= -L/usr/local/cuda/lib -lcudnn
else
CFLAGS+= -DCUDNN -I/opt/crc/c/cudnn/7.0.5/cuda/include
LDFLAGS+= -L/path/to/cudnn/<version>/cuda/lib64 -lcudnn
endif
endif

[Shruthi-Sampathkumar]

I am getting the same error. Did anybody solve it?

[SalahAdDin]

@Shruthi-Sampathkumar Use the fork, it worked for me.

[AyazSaiyed]

'/Users/yudiz/Downloads/darknet-master/build/darknet/x64/data/obj/000001.jpg'
It worked for me , issue was a extra 'space' between two new lines !!!

[amrkulk]

Use the strip() function in python to get rid of the trailing & leading new lines before feeding the image path for detection. The image paths in txt files contain extra new lines.

[pushgct]

The same happens here on ubuntu 18.04. Now here's the funny part: after retrying 5-10 times while randomly changing the file name it suddenly works. Even when I use files that previously gave an error. And once it works it remains working without any error.

Now I'm not really a linux expert, but could it be that there is some sort of background process that takes some time to start?

hi, which file did you rename to please? I too am stuck with same issue on Mac terminal.

[hxzmtk]

hi, you can use the python's imghdr to check the image Is it really jpg format

1. example

import imghdr
print(imghdr.what("/path/to/Object_Detection/Ground/JPEGImages/Training/6/um_50_na_sunny_sony_0_384.jpg"))
# maybe print  **webp* ! not jpg ~

1. now, write a sample filter script like this ~

import imghdr

def main():
    file_list = []
    with open("/path/to/Object_Detection/ground_train.txt") as f:
        for name in f.readlines():
            name = name.replace('\n','')
            try:
                img_type = name.split('.')[-1]
                if imghdr.what(name) in ['jpeg', img_type]:
                    print(name)
                    file_list.append(name)
            except FileNotFoundError:
                # print(name)
                continue
    with open("/path/to/Object_Detection/ground_train.txt", "w") as f:
        for name in file_list:
            f.write(name + '\n')

if __name__ == "__main__":
    main()