yeyupiaoling/pp-yoloe

PaddlePaddle实现的目标检测模型PP-YOLOE

开发者，你们好！

核心项目

项目类型	Pytorch版本	PaddlePaddle版本	备注
语音识别	MASR	PPASR
声纹识别	VoiceprintRecognition-Pytorch	VoiceprintRecognition-PaddlePaddle
声音分类	AudioClassification-Pytorch	AudioClassification-PaddlePaddle
语音情感识别	SpeechEmotionRecognition-Pytorch	SpeechEmotionRecognition-PaddlePaddle
语音合成	VITS-Pytorch	VITS-PaddlePaddle

语音项目

1. 基于PaddlePaddle动态图实现的语音识别项目：PPASR
2. 基于Pytorch实现的语音识别项目：MASR
3. 微调Whisper模型和加速推理：Whisper-Finetune
4. 基于PaddlePaddle静态图实现的语音识别项目：PaddlePaddle-DeepSpeech
5. 基于Pytorch实现的声音分类项目：AudioClassification-Pytorch
6. 基于PaddlePaddle实现声音分类项目：AudioClassification-PaddlePaddle
7. 基于PaddlePaddle实现声纹识别项目：VoiceprintRecognition-PaddlePaddle
8. 基于Pytorch实现声纹识别项目：VoiceprintRecognition-Pytorch
9. 基于Tensorflow实现声纹识别项目：VoiceprintRecognition-Tensorflow
10. 基于Keras实现声纹识别项目：VoiceprintRecognition-Keras
11. 基于PaddlePaddle实现的语音情感识别：SpeechEmotionRecognition-PaddlePaddle
12. 基于Pytorch实现的语音情感识别：SpeechEmotionRecognition-Pytorch
13. 基于PaddlePaddle实现的VIST语音合成：VITS-PaddlePaddle
14. 基于Pytorch实现的VIST语音合成：VITS-Pytorch

视觉项目

1. 基于PaddlePaddle实现的人脸识别项目：PaddlePaddle-MobileFaceNets
2. 基于Pytorch实现的人脸识别项目：Pytorch-MobileFaceNet
3. 基于PaddlePaddle实现的SSD目标检测模型：PaddlePaddle-SSD
4. 基于Pytorch实现的人脸关键点检测MTCNN模型：Pytorch-MTCNN
5. 基于PaddlePaddle实现的人脸关键点检测MTCNN模型：PaddlePaddle-MTCNN
6. 基于PaddlePaddle实现的文字识别CRNN模型：PaddlePaddle-CRNN
7. 基于PaddlePaddle实现的人流密度CrowdNet模型：PaddlePaddle-CrowdNet
8. 基于MXNET实现的年龄性别识别项目：Age-Gender-MXNET
9. 使用Tensorflow Lite、Paddle Lite、MNN、TNN框架在Android上不是图像分类模型：ClassificationForAndroid
10. 基于PaddlePaddle实现的PP-YOLOE模型：PP-YOLOE
11. 在Android部署的人脸检测、口罩识别、关键检测模型：FaceKeyPointsMask
12. 在Android上部署语义分割模型实现换人物背景：ChangeHumanBackground
13. 使用Tensorflow实现的人脸识别项目：Tensorflow-FaceRecognition

系列教程

1. PaddlePaddle V2版本系列教程：LearnPaddle
2. PaddlePaddle Fluid版本系列教程：LearnPaddle2

书籍源码

1. 《PaddlePaddle从入门到实战》源码：PaddlePaddleCourse
2. 《深度学习应用实战之PaddlePaddle》源码：BookSource

pp-yoloe's Issues

Error Broadcast dimension mismatch

Hi!

Thanks for your work, I have better performance with yoloe than yolo v5 and yolox no my dataset !
But i have a problem, on epoach 100 i get:

valueError: (InvalidArgument) Broadcast dimension mismatch. Operands could not be broadcast together with the shape of X = [16, 53, 2100] and the shape of Y = [16, 16, 2]. Received [53] in X is not equal to [16] in Y at i:1.
  [Hint: Expected x_dims_array[i] == y_dims_array[i] || x_dims_array[i] <= 1 || y_dims_array[i] <= 1 == true, but received x_dims_array[i] == y_dims_array[i] || x_dims_array[i] <= 1 || y_dims_array[i] <= 1:0 != true:1.] (at /paddle/paddle/phi/kernels/funcs/common_shape.h:84)

Error occour on ppyoloe_head.py:

assigned_labels, assigned_bboxes, assigned_scores = \
                self.assigner(
                pred_scores.detach(),
                pred_bboxes.detach() * stride_tensor,
                anchor_points,
                num_anchors_list,
                gt_labels,
                gt_bboxes,
                pad_gt_mask,
                bg_index=self.num_classes)

训练过程中出现问题

[2023-06-08 07:37:55.523808 INFO ] trainer:train:310 - Test epoch: 99, time/epoch: 0:04:26.419727, best_mAP: 0.83875, mAP: 0.81693
[2023-06-08 07:37:55.524025 INFO ] trainer:train:312 - ======================================================================
[2023-06-08 07:37:56.038528 INFO ] trainer:__save_checkpoint:196 - 已保存模型：models/PPYOLOE_M/epoch_99
Traceback (most recent call last):
File "train.py", line 44, in
trainer.train(num_epoch=args.num_epoch,
File "/data/dongxz/competion/cv/PP-YOLOE/ppyoloe/trainer.py", line 300, in train
self.__train_epoch(max_epoch=num_epoch, epoch_id=epoch_id, log_interval=log_interval, local_rank=local_rank, writer=writer)
File "/data/dongxz/competion/cv/PP-YOLOE/ppyoloe/trainer.py", line 204, in __train_epoch
output = self.model(data)
File "/data/anaconda3/envs/dongxz_paddlepaddle/lib/python3.8/site-packages/paddle/fluid/dygraph/layers.py", line 1012, in call
return self.forward(*inputs, **kwargs)
File "/data/dongxz/competion/cv/PP-YOLOE/ppyoloe/model/meta_arch.py", line 53, in forward
out = self.get_loss()
File "/data/dongxz/competion/cv/PP-YOLOE/ppyoloe/model/yolo.py", line 46, in get_loss
return self._forward()
File "/data/dongxz/competion/cv/PP-YOLOE/ppyoloe/model/yolo.py", line 36, in _forward
yolo_losses = self.yolo_head(neck_feats, self.inputs)
File "/data/anaconda3/envs/dongxz_paddlepaddle/lib/python3.8/site-packages/paddle/fluid/dygraph/layers.py", line 1012, in call
return self.forward(*inputs, **kwargs)
File "/data/dongxz/competion/cv/PP-YOLOE/ppyoloe/model/ppyoloe_head.py", line 202, in forward
return self.forward_train(feats, targets)
File "/data/dongxz/competion/cv/PP-YOLOE/ppyoloe/model/ppyoloe_head.py", line 142, in forward_train
return self.get_loss([
File "/data/dongxz/competion/cv/PP-YOLOE/ppyoloe/model/ppyoloe_head.py", line 308, in get_loss
self.assigner(
File "/data/anaconda3/envs/dongxz_paddlepaddle/lib/python3.8/site-packages/paddle/fluid/dygraph/layers.py", line 1012, in call
return self.forward(*inputs, **kwargs)
File "/data/anaconda3/envs/dongxz_paddlepaddle/lib/python3.8/site-packages/decorator.py", line 232, in fun
return caller(func, *(extras + args), **kw)
File "/data/anaconda3/envs/dongxz_paddlepaddle/lib/python3.8/site-packages/paddle/fluid/dygraph/base.py", line 375, in _decorate_function
return func(*args, **kwargs)
File "/data/dongxz/competion/cv/PP-YOLOE/ppyoloe/model/task_aligned_assigner.py", line 93, in forward
ious = iou_similarity(gt_bboxes, pred_bboxes)
File "/data/dongxz/competion/cv/PP-YOLOE/ppyoloe/model/bbox_utils.py", line 42, in iou_similarity
x2y2 = paddle.minimum(px2y2, gx2y2)
File "/data/anaconda3/envs/dongxz_paddlepaddle/lib/python3.8/site-packages/paddle/tensor/math.py", line 1008, in minimum
return _C_ops.minimum(x, y)
ValueError: (InvalidArgument) The 3-th dimension of input tensor is expected to be equal with the 3-th dimension of outputtensor 2 or 1, but received 0. (at /paddle/paddle/phi/kernels/funcs/broadcast_function.h:77)

oriented object detection

how can I implement OBB with your pp-yoloe? I can not find a solution about it. If you have any advice for this please tell me, thank you!

多卡训练该如何配置

export CUDA_VISIBLE_DEVICES=0,1 python3 -m paddle.distributed.launch --gpus 0,1 train.py --eval

执行上面的命令报错，麻烦看看，该如何调整支持多卡

VOC训练问题

您好，使用VOC数据按照教程训练时，出现ValueError: not enough values to unpack (expected 2, got 1)问题，请问怎么解决呢

dataset

怎么不能加载coco的官方预训练模型

模型训练很慢是什么问题呀？

How to do inference with batch_size > 1 ?

您好, 目前使用infer.py可以進行batch_size=1的預測
請問有辦法使其進行batch_size>1的預測嗎？
謝謝

运行train.py报错

Error: Cannot load cudnn shared library.

how to slove the problem dispalyed bellow?
Error: Cannot load cudnn shared library. Cannot invoke method cudnnGetVersion at (/paddle/paddle/fluid/platform/dynload/cudnn.cc:63)

CUDA_VISIBLE_DEVICES 无法指定 GPU

我使用 CUDA_VISIBLE_DEVICES=3 python train.py 后，仍然会判断 GPU 0 的占用情况，无法用其他卡训练。请问怎么指定 GPU 进行训练呢？报错如下（GPU0正在使用）

Out of memory error on GPU 0. Cannot allocate 27.000000MB memory on GPU 0, 10.744995GB memory has been allocated and available memory is only 16.562500MB.

Please check whether there is any other process using GPU 0.

1. If yes, please stop them, or start PaddlePaddle on another GPU.
2. If no, please decrease the batch size of your model.
   If the above ways do not solve the out of memory problem, you can try to use CUDA managed memory. The command is export FLAGS_use_cuda_managed_memory=false.
   (at /paddle/paddle/fluid/memory/allocation/cuda_allocator.cc:87)
   . (at /paddle/paddle/fluid/imperative/tracer.cc:307)

2020-10-21 22:58:00,947-WARNING: Found an invalid bbox in annotations: xml_file: dataset/annotation/001113.xml, x1: 3765.0, y1: 1851.0, x2: 3765.0, y2: 1871.0.

2020-10-21 22:58:00,947-WARNING: Found an invalid bbox in annotations: xml_file: dataset/annotation/001113.xml, x1: 3765.0, y1: 1851.0, x2: 3765.0, y2: 1871.0.

训练中断

您好，我在用voc格式的数据集训练中每迭代一千次左右（每次中断时间不同）就会报错：

2020-10-02 20:42:48,441-WARNING: fail to map op [Gt2YoloTarget_82e929] with error: index 18 is out of bounds for axis 2 with size 16 and stack:
Traceback (most recent call last):
File "C:\Users\LHT\Anaconda3\envs\paddle\lib\site-packages\ppdet\data\reader.py", line 46, in call
data = f(data, ctx)
File "C:\Users\LHT\Anaconda3\envs\paddle\lib\site-packages\ppdet\data\transform\batch_operators.py", line 258, in call
target[best_n, 0, gj, gi] = gx * grid_w - gi
IndexError: index 18 is out of bounds for axis 2 with size 16
每次会中断训练，重新开始训练会在不同的位置出现这个问题。