训练时 AP AR 一直为0
WjzZwd opened this issue · comments
我制作了自己的数据集 并且整理为了coco格式
同时我修改了https://github.com/lyuwenyu/RT-DETR/blob/main/rtdetr_pytorch/src/data/coco/coco_dataset.py#L154处为我的类
并且相对应地修改了coco_detection.yaml中的代码
但训练时结果一直如下图所示
请问这是为什么呢
(ps:不知道这个warning有没有影响
warning没影响, 可视化看下是框不对还是类别不对
我的问题解决了,这是因为我在制作COCO数据集时,annotations中的每一个元素的id,我都会在下一个图片中重新从1初始化开始,实际上,这个id应该是贯彻了整个annotations的,不应该在每个新的图片来到时初始化
Hello @WjzZwd , can I ask you something about RTDETR. I have train RT DETR using an aquarium dataset in Roboflow but the result is very bad. Can you provide the dataset you use to train and some config for me. Thanks Vào Th 4, 20 thg 3, 2024 vào lúc 09:27 windy_z @.> đã viết:
…
which file you want? ---原始邮件--- 发件人: @.> 发送时间: 2024年3月20日(周三) 上午10:25 收件人: @.>; 抄送: @.@.>; 主题: Re: [lyuwenyu/RT-DETR] 训练时 AP AR 一直为0 (Issue #232) @WjzZwd Can you provide your new config — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.> — Reply to this email directly, view it on GitHub <#232 (comment)>, or unsubscribe https://github.com/notifications/unsubscribe-auth/AVJNOIOM6DU3AVVMDIY3X7TYZDXXZAVCNFSM6AAAAABE2EABKGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDAMBYGU3TINRXGE . You are receiving this because you commented.Message ID: @.***>
我只修改了这里 但我认为这个文件被是否被修改都没关系,因为我在下图中 将remap_mscoco_category:设置为了False
Can you provide your email. I want ask you something
Can I ask you something. I use aquarium dataset from roboflow(https://public.roboflow.com/object-detection/aquarium) . Config I capture in image attached. I train many epoch but ap is very low and loss is very high(approximately 20). Can you provide your config and data you use. @WjzZwd @lyuwenyu
below is part of file
"annotations": [
{
"id": 1,
"image_id": 1,
"category_id": 1,
"iscrowd": 0,
"area": 507.4649428747899,
"bbox": [
626.001225,
336.593144,
15.554194000000052,
32.625602000000015
]
},
{
"id": 2,
"image_id": 1,
"category_id": 1,
"iscrowd": 0,
"area": 454.49715203815197,
"bbox": [
635.480478,
259.430917,
14.563128000000006,
31.208758999999986
]
},
{
"id": 3,
"image_id": 1,
"category_id": 1,
"iscrowd": 0,
"area": 539.1054014736708,
"bbox": [
1040.808555,
50.681948,
20.009428999999955,
26.942568
]
},
{
"id": 4,
"image_id": 1,
"category_id": 1,
"iscrowd": 0,
"area": 729.635269491221,
"bbox": [
1005.685953,
251.969635,
23.48572499999989,
31.067180999999977
]
},
{
"id": 5,
"image_id": 1,
"category_id": 1,
"iscrowd": 0,
"area": 357.6577425163594,
"bbox": [
569.194116,
59.548848,
13.17986899999994,
27.136669000000005
]
},
{
"id": 6,
"image_id": 1,
"category_id": 1,
"iscrowd": 0,
"area": 317.45542948531806,
"bbox": [
659.241073,
29.226642,
11.993859999999927,
26.468162
]
},
{
"id": 7,
"image_id": 1,
"category_id": 1,
"iscrowd": 0,
"area": 371.71030207648755,
"bbox": [
779.401501,
17.402781,
14.185329999999908,
26.203853
]
},
{
"id": 8,
"image_id": 1,
"category_id": 1,
"iscrowd": 0,
"area": 425.3477583848202,
"bbox": [
445.691356,
60.188679,
15.66622000000001,
27.150630999999997
]
},
{
"id": 9,
"image_id": 1,
"category_id": 2,
"iscrowd": 0,
"area": 3331.425140328181,
"bbox": [
706.041467,
658.750551,
55.29386899999997,
60.24944900000003
]
},
"images": [
{
"id": 1,
"width": 1280,
"height": 720,
"file_name": "0_000000.png"
},
{
"id": 2,
"width": 1280,
"height": 720,
"file_name": "0_000001.png"
},
{
"id": 3,
"width": 1280,
"height": 720,
"file_name": "0_000003.png"
},
{
"id": 4,
"width": 1280,
"height": 720,
"file_name": "0_000004.png"
},
{
"id": 5,
"width": 1280,
"height": 720,
"file_name": "0_000005.png"
},
{
"id": 6,
"width": 1280,
"height": 720,
"file_name": "0_000006.png"
},
{
"id": 7,
"width": 1280,
"height": 720,
"file_name": "0_000008.png"
},
{
"id": 8,
"width": 1280,
"height": 720,
"file_name": "0_000012.png"
},
{
"id": 9,
"width": 1280,
"height": 720,
"file_name": "0_000013.png"
},
{
"id": 10,
"width": 1280,
"height": 720,
"file_name": "0_000015.png"
},
{
"id": 11,
"width": 1280,
"height": 720,
"file_name": "0_000016.png"
},
{
"id": 12,
"width": 1280,
"height": 720,
"file_name": "0_000018.png"
},
{
"id": 13,
"width": 1280,
"height": 720,
"file_name": "0_000021.png"
},
{
"id": 14,
"width": 1280,
"height": 720,
"file_name": "0_000022.png"
},
{
"id": 15,
"width": 1280,
"height": 720,
"file_name": "0_000023.png"
},
{
"id": 16,
"width": 1280,
"height": 720,
"file_name": "0_000024.png"
},
{
"id": 17,
"width": 1280,
"height": 720,
"file_name": "0_000025.png"
},
{
"id": 18,
"width": 1280,
"height": 720,
"file_name": "0_000026.png"
},
我的问题解决了,这是因为我在制作COCO数据集时,annotations中的每一个元素的id,我都会在下一个图片中重新从1初始化开始,实际上,这个id应该是贯彻了整个annotations的,不应该在每个新的图片来到时初始化
能详细说说吗,我遇到了相似的问题,几个简化版的coco数据集在DAB-DETR Achors-detr都出现了ap一直为0的问题
我已为此困扰多天,希望您能帮帮我