profile
viewpoint

zxj11838/TensorFlow-Examples 1

TensorFlow Tutorial and Examples for Beginners with Latest APIs

zxj11838/tensorflow 0

Computation using data flow graphs for scalable machine learning

issue openedultralytics/yolov5

j, k = ((gxy % 1. < g) & (gxy > 1.)).T

❔Question

In the build_target() method, when calculating j、k, Does it need to be written as follows:

j, k = ((gxy % 1. < g) & (gxy > 1.) & (gxy < torch.tensor(p[i].shape)[2] - 1)).T

Additional context

created time in 4 hours

issue openedultralytics/yolov5

Objectness loss

❔Question

In the compute_loss() method, when calculating the confidence loss, only calculate the confidence loss of the positive sample? Did the negative sample not participate in the calculation?

Additional context

tobj[b, a, gj, gi] = (1.0 - model.gr) + model.gr * iou.detach().clamp(0).type(tobj.dtype)

created time in 3 days

startedWZTENG/YOLOv5_NCNN

started time in 10 days

issue commentsunnyden/YOLOV5_NCNN_Android

how to replaced the input with a downscaled image and stacked it to match the channel number?

@Skr20200701 I used your method and succeeded, but the coordinates of the box seem to be reduced by half. I multiply w and h by 2 in decode_infer() in the YoloV5.cpp file, and the box is displayed correctly. Is this correct?

image

panchengl

comment created time in 11 days

issue commentsunnyden/YOLOV5_NCNN_Android

检测失效

The target can be detected on the android side, but the coordinates are not accurate. Has anyone encountered this situation?

Dreamskya

comment created time in 11 days

issue openedsunnyden/YOLOV5_NCNN_Android

ncnn model

In this project, you need to specify the output of the model in YoloV5.h image

created time in 11 days

issue commentdog-qiuqiu/MobileNet-Yolo

训练自己的数据集

使用MobileNetV2-YOLOv3-Lite这个网络训练自己的数据集,出现如下信息: CUDA-version: 9010 (10000), cuDNN: 7.5.0, GPU count: 2 OpenCV isn't used - data increase will run slowly MobileNetV2-YOLOv3-Lite-voc 段错误 (核心已转储)

有人知道如何解决吗?

liuwei3612

comment created time in 11 days

issue commentdog-qiuqiu/Yolo-Fastest

请问如果我想训练自己的数据,需要怎么更改.cfg文件?

You can refer to this part of (How to train )the project(https://github.com/AlexeyAB/darknet).

JimXu1989

comment created time in 12 days

issue openeddog-qiuqiu/Android_NCNN_yolov4-tiny

Model output

std::vector<BoxInfo> YoloV5::detect(JNIEnv* env, jobject image) In this method, these two parameters (threshold, nms_threshold)are not used. In the param file,Is the decoding of boxes and nms operation included in the model?

created time in a month

issue openeddog-qiuqiu/Yolo-Fastest

Android deployment

Can I use this project(Android_NCNN_yolov4-tiny) to deploy?

created time in a month

issue commentsunnyden/YOLOV5_NCNN_Android

how to replaced the input with a downscaled image and stacked it to match the channel number?

@ shengyucaihua Did you solve it? The detected boxes is still incorrect.

panchengl

comment created time in a month

issue commentWZTENG/YOLOv5_NCNN

yolo4和yolo5进行预测,均检测不到任何目标

@xinsuinizhuan 我也是检测不到任何目标,麻烦问一下,你是怎么解决的?

xinsuinizhuan

comment created time in a month

starteddog-qiuqiu/Yolo-Fastest

started time in a month

issue openedTianxiaomo/pytorch-YOLOv4

coco.names training/testing

I want to know where to set voc.names and coco.names when training my own data set? Or just need to set this file when testing?

created time in a month

issue commentTianxiaomo/pytorch-YOLOv4

How to prepare 'val.txt' on our own dataset?

You can refer to this file (Use_yolov4_to_train_your_own_data.md)in the project.

jayhung97724

comment created time in a month

issue commentTianxiaomo/pytorch-YOLOv4

NotImplementedError: Create your own 'get_image_id' function

@weilanShi I have sovled it,thanks.

yuxx0218

comment created time in a month

issue closedTianxiaomo/pytorch-YOLOv4

Training on VOC datasets

Does it support training on VOC dataset? When I train, I get an error, it seems to be a problem with the dataset. Has anyone successfully trained on the VOC dataset?

closed time in a month

zxj11838

issue commentTianxiaomo/pytorch-YOLOv4

我有几个疑问

@ oreo-lp

Loss includes confidence loss, bounding box loss, and category loss.

obj_mask: which is the positive and negative sample mask after screening,used to calculate the confidence loss;

tgt_mask: which is the mask of positive samples, used to calculate the bounding box loss and category loss;

tgt_scale: used to balance the loss of large and small targets. Avoid that the error of the large frame has a greater impact on the loss than the error of the small frame.

shanniruo

comment created time in a month

issue commentultralytics/yolov5

Convert pytorch model to ncnn format

According to your method, I successfully converted the ncnn model. When testing in the Android project, many boxes appeared, and the accuracy rate was very low. Does anyone know how to solve it?

zxj11838

comment created time in 2 months

issue commentultralytics/yolov5

Convert pytorch model to ncnn format

Thank you for your answer, I would like to know whether it will affect the effect of the model?

zxj11838

comment created time in 2 months

startedxiangweizeng/darknet2ncnn

started time in 2 months

issue openedultralytics/yolov5

Convert pytorch model to ncnn format

❔Question

I try to convert pytorch model to ncnn format,the conversion to onnx format is successful. When converting from onnx to ncnn, the following prompt appears: Unsupported slice step ! Unsupported slice step ! Has anyone encountered this problem?

Additional context

created time in 2 months

starteddog-qiuqiu/MobileNet-Yolo

started time in 2 months

startedAlexeyAB/darknet

started time in 2 months

more