PyTorch-YOLOv3

PyTorch-YOLOv3 copied to clipboard

Does not give the exact same result as darknet

1

Using the exact same image size (416x416), confidence threshold, nms threshold, config file (yolov3.cfg) and weights file (yolov3.weights), inferring on dog.jpg gives different results when using this repo and darknet....

pfeatherstone

not enough values to unpack (expected 2, got 1)

2

python3 train.py --data_config config/yolov3-tiny.cfg --pretrained_weights weights/yolov3-tiny.weights Traceback (most recent call last): File "train.py", line 50, in data_config = parse_data_config(opt.data_config) File "/home/ubuntu/myfile/project/Pytorch_YoloV3/PyTorch-YOLOv3/utils/parse_config.py", line 34, in parse_data_config key, value = line.split('=') ValueError:...

Alexyitx

learning_rate not being read in/config not truly being parsed

1

As far as I can tell the learning_rate and decay are not actually being read in from the config file to tweak the adam optimizer. Despite changing the learning rate...

jshermeyer

custom dataset examples with no bounding boxes?

2

Is it not possible to give the network some examples with no bounding boxes for training? I tried using empty text files for the labels but it crashes.

dasturge

mrright2019

<11% mAP with pretrained YOLO after tiny amount of training

1

Only 200 iterations of training on COCO completely destroys the accuracy of the pretrained YOLOv3 net. I am using an unchanged repo with the COCO dataset downloaded using the script...

half-potato

when training with custom data, got this error during evaluate: Traceback (most recent call last): File "train.py", line 169, in batch_size=8, File "/data/PyTorch-YOLOv3/test.py", line 48, in evaluate outputs = non_max_suppression(outputs,...

turboxin

what's the yolo[i] mean in the results when i trained??

1

yhx0105

I would like some more information on the yolov3.weights pretrained weights. Were they learned from trianing on 100% of the 2017 coco training data? Or was some data held back...

astrobdr

Hello, I would like to ask why I use yolov3.weights for training, the effect is the same as that of using darknet53.conv.74 for initialization. Shouldn't yolov3.weights for initialization of loss...

Gmy12138