使用ipdb调试

try:
import ipdb
except:
import pdb as ipdb ipdb.set_trace()

测试inference:

# coding=utf-8

import matplotlib.pyplot as plt
import matplotlib.pylab as pylab import requests
from io import BytesIO
from PIL import Image
import numpy as np # this makes our figures bigger
pylab.rcParams['figure.figsize'] = 20, 12 from maskrcnn_benchmark.config import cfg
from predictor import COCODemo config_file = "../configs/caffe2/e2e_mask_rcnn_R_50_FPN_1x_caffe2.yaml"
#config_file = "../configs/e2e_mask_rcnn_R_50_FPN_1x.yaml" # update the config options with the config file
cfg.merge_from_file(config_file)
# manual override some options
cfg.merge_from_list(["MODEL.DEVICE", "cuda"]) # only "cuda" and "cpu" are valid device types
coco_demo = COCODemo(
cfg,
min_image_size=800,
confidence_threshold=0.7,
) def load(url):
"""
Given an url of an image, downloads the image and
returns a PIL image
"""
response = requests.get(url)
pil_image = Image.open(BytesIO(response.content)).convert("RGB")
# convert to BGR format
image = np.array(pil_image)[:, :, [2, 1, 0]]
return image def imshow(img):
plt.imshow(img[:, :, [2, 1, 0]])
plt.axis("off")
plt.show() # from http://cocodataset.org/#explore?id=345434
image = load("http://farm3.staticflickr.com/2469/3915380994_2e611b1779_z.jpg")
# image = Image.open("474797538.jpg").convert("RGB")
# image = np.array(image)[:, :, [2, 1, 0]] #imshow(image) # compute predictions
predictions = coco_demo.run_on_opencv_image(image)
imshow(predictions)

在predictor.py文件中核心函数def compute_prediction(self, original_image):下的变量信息:

->输入original_image=[480,640,3],int整型数据;

->经过变换后image=[3,800,1066],数据torch.float32;

然后进入核心函数:predictions = self.model(image_list),跳入generalized_rcnn.py文件,中def forward(self, images, targets=None):函数;

经过features = self.backbone(images.tensors)函数,使用各种基网络(如ResNet-50_FPN)提取各个stage的特征图;然后使用feature map进行RPN及ROI pooling操作;

-> features变量信息,tuple类型,5个特征图的tensor:

ipdb> p features.size()
*** AttributeError: 'tuple' object has no attribute 'size'
ipdb> p features.shape()
*** AttributeError: 'tuple' object has no attribute 'shape'
ipdb> p features[0].shape()
*** TypeError: 'torch.Size' object is not callable
ipdb> p features[0].size()
torch.Size([1, 256, 200, 272])
ipdb> p features[1].size()
torch.Size([1, 256, 100, 136])
ipdb> p features[2].size()
torch.Size([1, 256, 50, 68])
ipdb> p features[3].size()
torch.Size([1, 256, 25, 34])
ipdb> p features[4].size()
torch.Size([1, 256, 13, 17])

->经过rpn网络得到候选框:proposals, proposal_losses = self.rpn(images, features, targets)

ipdb> targets
ipdb> p targets
None
ipdb> p images
<maskrcnn_benchmark.structures.image_list.ImageList object at 0x7f5128049f28>
ipdb> p proposal_losses
{}
ipdb> p proposals
[BoxList(num_boxes=1000, image_width=1066, image_height=800, mode=xyxy)]

-> 然后经过fast rcnn网络,x, result, detector_losses = self.roi_heads(features, proposals, targets); 这部分有在roi_heads.py文件中,由两分支组成:检测分支和分割分支组成;

->在roi_heads.py文件的forward()中:x, detections, loss_box = self.box(features, proposals, targets)得到检测结果,

    def forward(self, features, proposals, targets=None):
"""
Arguments:
features (list[Tensor]): feature-maps from possibly several levels
proposals (list[BoxList]): proposal boxes
targets (list[BoxList], optional): the ground-truth targets. Returns:
x (Tensor): the result of the feature extractor
proposals (list[BoxList]): during training, the subsampled proposals
are returned. During testing, the predicted boxlists are returned
losses (dict[Tensor]): During training, returns the losses for the
head. During testing, returns an empty dict.
""" if self.training:
# Faster R-CNN subsamples during training the proposals with a fixed
# positive / negative ratio
with torch.no_grad():
proposals = self.loss_evaluator.subsample(proposals, targets) # extract features that will be fed to the final classifier. The
# feature_extractor generally corresponds to the pooler + heads
x = self.feature_extractor(features, proposals)
# final classifier that converts the features into predictions
class_logits, box_regression = self.predictor(x) if not self.training:
result = self.post_processor((class_logits, box_regression), proposals)
return x, result, {} loss_classifier, loss_box_reg = self.loss_evaluator(
[class_logits], [box_regression]
)
return (
x,
proposals,
dict(loss_classifier=loss_classifier, loss_box_reg=loss_box_reg),
)

->x为经过池化操作及特征提取的特征用于分类回归,经过后处理,剩下有用的box返回;

->筛选出来的1000个proposals,提取1024维特征; 最终有效box剩88个;

ipdb> x.shape
torch.Size([1000, 1024])
ipdb> detections.shape
*** AttributeError: 'list' object has no attribute 'shape'
ipdb> detections.size()
*** AttributeError: 'list' object has no attribute 'size'
ipdb> len(detections)
1
ipdb> detections
[BoxList(num_boxes=88, image_width=1066, image_height=800, mode=xyxy)]

-> 利用检测的结果,经过mask分支:x, detections, loss_mask = self.mask(mask_features, detections, targets); mask分支:

    def forward(self, features, proposals, targets=None):
"""
Arguments:
features (list[Tensor]): feature-maps from possibly several levels
proposals (list[BoxList]): proposal boxes
targets (list[BoxList], optional): the ground-truth targets. Returns:
x (Tensor): the result of the feature extractor
proposals (list[BoxList]): during training, the original proposals
are returned. During testing, the predicted boxlists are returned
with the `mask` field set
losses (dict[Tensor]): During training, returns the losses for the
head. During testing, returns an empty dict.
""" if self.training:
# during training, only focus on positive boxes
all_proposals = proposals
proposals, positive_inds = keep_only_positive_boxes(proposals)
if self.training and self.cfg.MODEL.ROI_MASK_HEAD.SHARE_BOX_FEATURE_EXTRACTOR:
x = features
x = x[torch.cat(positive_inds, dim=0)]
else:
x = self.feature_extractor(features, proposals)
mask_logits = self.predictor(x) if not self.training:
result = self.post_processor(mask_logits, proposals)
return x, result, {} loss_mask = self.loss_evaluator(proposals, mask_logits, targets) return x, all_proposals, dict(loss_mask=loss_mask)

->x为maks分支特征的tensor,变成[88, 256, 14, 14],返回的detections就是box+mask的内容

ipdb> x.shape
torch.Size([88, 256, 14, 14])
ipdb> detections
[BoxList(num_boxes=88, image_width=1066, image_height=800, mode=xyxy)]
ipdb> loss_mask
{}

->做完后,返回generalized_rcnn.py文件,返回predictor.py进行一些后处理,可视化结果即可!

maskrcnn_benchmark代码分析(1)的更多相关文章

  1. maskrcnn_benchmark代码分析(2)

    maskrcnn_benchmark训练过程 ->训练命令: python tools/train_net.py --config-file "configs/e2e_mask_rcn ...

  2. maskrcnn_benchmark代码分析(3)

    数据结构 数据加载 数据后处理

  3. Android代码分析工具lint学习

    1 lint简介 1.1 概述 lint是随Android SDK自带的一个静态代码分析工具.它用来对Android工程的源文件进行检查,找出在正确性.安全.性能.可使用性.可访问性及国际化等方面可能 ...

  4. pmd静态代码分析

    在正式进入测试之前,进行一定的静态代码分析及code review对代码质量及系统提高是有帮助的,以上为数据证明 Pmd 它是一个基于静态规则集的Java源码分析器,它可以识别出潜在的如下问题:– 可 ...

  5. [Asp.net 5] DependencyInjection项目代码分析-目录

    微软DI文章系列如下所示: [Asp.net 5] DependencyInjection项目代码分析 [Asp.net 5] DependencyInjection项目代码分析2-Autofac [ ...

  6. [Asp.net 5] DependencyInjection项目代码分析4-微软的实现(5)(IEnumerable<>补充)

    Asp.net 5的依赖注入注入系列可以参考链接: [Asp.net 5] DependencyInjection项目代码分析-目录 我们在之前讲微软的实现时,对于OpenIEnumerableSer ...

  7. 完整全面的Java资源库(包括构建、操作、代码分析、编译器、数据库、社区等等)

    构建 这里搜集了用来构建应用程序的工具. Apache Maven:Maven使用声明进行构建并进行依赖管理,偏向于使用约定而不是配置进行构建.Maven优于Apache Ant.后者采用了一种过程化 ...

  8. STM32启动代码分析 IAR 比较好

    stm32启动代码分析 (2012-06-12 09:43:31) 转载▼     最近开始使用ST的stm32w108芯片(也是一款zigbee芯片).开始看他的启动代码看的晕晕呼呼呼的. 还好在c ...

  9. 常用 Java 静态代码分析工具的分析与比较

    常用 Java 静态代码分析工具的分析与比较 简介: 本文首先介绍了静态代码分析的基 本概念及主要技术,随后分别介绍了现有 4 种主流 Java 静态代码分析工具 (Checkstyle,FindBu ...

随机推荐

  1. 基于SOA的银行系统架构

    Part-1  [简述] 1.通过引入面向服务架构(SOA),企业服务总线(ESB),适配器(Adapter)及面向构件等技术,尝试打造一个统一业务流程服务平台,实现面向流程的服务集成. 2.传统银行 ...

  2. 子类 父类强转 HttpServlet service实现

    相当于 走父类 临时走了一趟 HttpServletRequest ->ServletRequets -> HttpServeltRequest /* */ public void ser ...

  3. 【Codeforces666E】Forensic Examination 后缀自动机 + 线段树合并

    E. Forensic Examination time limit per test:6 seconds memory limit per test:768 megabytes input:stan ...

  4. Atcoder Grand Contest 010 B - Boxes 差分

    B - Boxes 题目连接: http://agc010.contest.atcoder.jp/tasks/agc010_b Description There are N boxes arrang ...

  5. WorldFinal11 (2/11)

    WorldFinal 11 Trash Removal 题意 给你一个多边形,问这个多边形至少需要多宽的长度,才能把这个多边形放进去. 数据范围100 题解 数据范围只有100,暴力枚举两点,然后算最 ...

  6. GitLab查询当前版本

    gitlab-rake gitlab:env:info 其实还有很多方法可以参考GitLab的帮助文档:https://docs.gitlab.com/omnibus/README.html 参考: ...

  7. 【 D3.js 入门系列 --- 9 】 常见可视化图形

    本人的个人博客为: www.ourd3js.com csdn博客为: blog.csdn.net/lzhlzz 转载请注明出处,谢谢. Layout ,直译为"布局,安排".但在 ...

  8. Bitbox : a small open, DIY 32 bit VGA console

    Bitbox : a small open, DIY 32 bit VGA console Hi all, I've been developing a simple DIY console and ...

  9. PG的集群技术:Pgpool-II与Postgres-XC Postgres-XL Postgres-XZ Postges-x2

    https://segmentfault.com/a/1190000007012082 https://www.postgres-xl.org/ https://www.biaodianfu.com/ ...

  10. Groovy中Closure的this到底指向谁?

    Groovy in Action(中文版)第136页明确说Closure的this指向Closure自己.并且从代码注释处作者也是这样理解的: class Mother{    int field = ...