Segmentation fault (core dumped) in training own data yolo2 darknet

[aminullah6264]

i have prepare my data according to the instruction given in this link

https://pjreddie.com/darknet/yolo/

i have downloaded weights also. i have 12GB titanX gpu when i run the darknet for training it give this error. i use following line for training ./darknet detector train cfg/voc.data cfg/yolo-voc.cfg darknet19_448.conv.23

plz how to solve this problem or why it is comming

[aminullah6264]

And When i try to run with tiny yolo it return this

[workingforfood]

Did you run the examples? Do they work correctly for you? If it is okay for examples, then can you share your cfg/voc.data cfg/yolo-voc.cfg

[aminullah6264]

Dear same thing is coming i run example. This screenshots are same when i run examples.

[rperrones]

Enable debug option at Makefile and compile source code again. run darknet in gdb to be able to trace segment fault. The 'run/backtrace/where' commands probably will point the line that rises the fault.

[Dmitrivm]

Aminullah6264, show your train list file.

[lo-pan]

Do you solve this problems now? I also have the same problem in v3.Now i can't train my data,so i hope you can help me slove this probel,thanks.

[ahsan856jalal]

It will be great if you show us your cfg file

[dee6600]

`[net]

Testing

batch=24 subdivisions=8

Training

batch=64

subdivisions=8

width=416 height=416 channels=3 momentum=0.9 decay=0.0005 angle=0 saturation = 1.5 exposure = 1.5 hue=.1

learning_rate=0.001 burn_in=1000 max_batches = 500200 policy=steps steps=400000,450000 scales=.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

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

[reorg] stride=2

[route] layers=-1,-4

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

[convolutional] size=1 stride=1 pad=1 filters=40 activation=linear

[region] anchors = 0.57273, 0.677385, 1.87446, 2.06253, 3.33843, 5.47434, 7.88282, 3.52778, 9.77052, 9.16828 bias_match=1 classes=3 coords=4 num=5 softmax=1 jitter=.3 rescore=1

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

absolute=1 thresh = .6 random=1`

i am facing the same issue of seg fault. this cfg is for 3 classes. i have gtx860 4gb. help plz

[Jerry3062]

i am facing the same issue of segmentation fault. with tried many solutions,the question not solved.finally,i change my label coordinate which is zero to a tiny float,and it's work.i think i will help for somebody.ignore my bad english.

[saivineethkumar]

@Jerry3062 can you explain a bit clearly what to change. what do you mean by label coordinate and in which file it is present? thanks!

[arun-kumark]

Hi saivineethkumar, in label.txt, I found some coordinates and were zero (0.0). Remove those reference from the file list. Its running without any error so far.

[singhnarotam1997]

Change random flag in the last line of cfg file to 0. Core is getting dumped because image is being resized to very high dimension after some iterations(608 in your case) taking too much memory. If you want random dimensions to increase precision, maybe run the model on cpu instead of gpu.

[Jerry3062]

@saivineethkumar The annotation file. coordinate means (x,y,w,h) or (x1,y1,x2,y2),i forget yolo3's format. In my dataset,some x or y is zero value.

[Jibin-John]

where present label.txt

[sachindesh]

check out my comment, if it helps; https://github.com/pjreddie/darknet/issues/174#issuecomment-445203621

[wilkice]

1. The issue with 'cannot load images', 'segmentation fault (core dump)', 'cannot fopen', 'cannot open label file', is that the files edited in Windows or any operating system that doesn't support Unix style file formats ('\r' line ending) are transferred to Unix boxes (Ubuntu 16 in my case).
2. dos2unix, "tr -d '\r' < file > file" tools used on Ubuntu on txt as well as JPG files, but it doesn't work even. Solution Whatever editing/saving of image files, txt files or any other files, including the marking of objects (yolo_mark tool) should be done only using the Ubuntu or like desktops and not on Windows or non-Unix style operating systems. Cheer!!

check out my comment, if it helps; #174 (comment)

This problem solved after I changed yolov3.weights to yolov3-tiny.weights, also I changed yolov3.cfg to yolov3-tiny.cfg. Because my GPU is only 1G memory but yolov3 needs 4G. So if your GPU's memory is lower than 4G, you can try yolov3-tiny

[Mahibro]

Hey guys, Here am training on small dataset with larger size(3520*4280) with yolov3-tiny & darknet19_448.conv.23 , even i'm facing the same issue(Segmentation fault-core dumped),i made changes in configuration file(random 1 to 0, batch & Subdivision).Can somebody help me to resolve this?

[Sharev]

In my case my training data was the culprit. Make sure your training data is correct. Specifically, I removed a class after using it on few of the images, which raised this issue.

[Rapternmn]

Use AlexeyAB repo for better exception handling. Some of your data in annotations file might be going out of bound (x, y < 0 / > 1 )

[afcarvalho1991]

For future reference...

corrupted images can also cause a segmentation fault (core dumped) during training (and probably also during detection!).

In my case, after a few iterations (with no clear pattern) training would just halt and output segmentation fault (core dumped). Hope it helps!

Best regards, André

[pandadom]

Had segmentation fault, and found that they were caused by objects that are partially or fully outside of the image. Some of them were caught by yolo and listed as "bad-label" but some of them i had to identify myself. After removing them from the data-set the training succeeds!

[rumanxyz]

I was having this very similar problem while fine tuning YoloV4. Issue was with the data. .ipynb_checkpoints accidentally got into the train.txt. So I would recommend you to also once look at the images in train.txt file.

I was getting Segmentation fault