diff --git a/.gitignore b/.gitignore index 3fd7e4f..579edac 100644 --- a/.gitignore +++ b/.gitignore @@ -4,3 +4,6 @@ build/ *.pth* *.pkl lf-net-release/release/models +data +r2d2 +log \ No newline at end of file diff --git a/.gitmodules b/.gitmodules new file mode 100644 index 0000000..7ecfbbf --- /dev/null +++ b/.gitmodules @@ -0,0 +1,3 @@ +[submodule "r2d2"] + path = r2d2 + url = git@github.com:changhaonan/r2d2.git diff --git a/.vscode/settings.json b/.vscode/settings.json deleted file mode 100644 index 03f58a1..0000000 --- a/.vscode/settings.json +++ /dev/null @@ -1,105 +0,0 @@ -{ - "files.associations": { - "*.cu": "cpp", - "*.cuh": "cpp", - "*.vert": "cpp", - "*.frag": "cpp", - "*.glsl": "cpp", - "*.geom": "cpp", - "*.urdf": "xml", - "*.xacro": "xml", - "*.dae": "xml", - "*.pyx": "python", - "*.pxd": "cpp", - "*.srdf": "xml", - "*.sdf": "xml", - "array": "cpp", - "bitset": "cpp", - "string_view": "cpp", - "initializer_list": "cpp", - "regex": "cpp", - "utility": "cpp", - "limits": "cpp", - "atomic": "cpp", - "*.tcc": "cpp", - "cctype": "cpp", - "cmath": "cpp", - "complex": "cpp", - "cstddef": "cpp", - "cstdint": "cpp", - "cstdio": "cpp", - "cstdlib": "cpp", - "cwchar": "cpp", - "deque": "cpp", - "list": "cpp", - "unordered_map": "cpp", - "vector": "cpp", - "exception": "cpp", - "algorithm": "cpp", - "functional": "cpp", - "tuple": "cpp", - "type_traits": "cpp", - "iosfwd": "cpp", - "iostream": "cpp", - "istream": "cpp", - "memory": "cpp", - "new": "cpp", - "ostream": "cpp", - "numeric": "cpp", - "stdexcept": "cpp", - "streambuf": "cpp", - "thread": "cpp", - "cinttypes": "cpp", - "typeinfo": "cpp", - "iterator": "cpp", - "map": "cpp", - "memory_resource": "cpp", - "random": "cpp", - "string": "cpp", - "core": "cpp", - "numericaldiff": "cpp", - "graphicscontext": "cpp", - "quat": "cpp", - "vec2b": "cpp", - "vec2d": "cpp", - "vec2f": "cpp", - "vec2i": "cpp", - "vec2s": "cpp", - "vec2ub": "cpp", - "vec2ui": "cpp", - "vec2us": "cpp", - "vec3b": "cpp", - "vec3d": "cpp", - "vec3f": "cpp", - "vec3i": "cpp", - "vec3s": "cpp", - "vec3ub": "cpp", - "vec3ui": "cpp", - "vec3us": "cpp", - "vec4b": "cpp", - "vec4d": "cpp", - "vec4f": "cpp", - "vec4i": "cpp", - "vec4s": "cpp", - "vec4ui": "cpp", - "vec4us": "cpp", - "viewport": "cpp", - "*.ipp": "cpp", - "camera": "cpp", - "plane": "cpp", - "polytope": "cpp", - "export": "cpp", - "optional": "cpp", - "variant": "cpp", - "mixinvector": "cpp", - "fast_back_stack": "cpp", - "nonlinearoptimization": "cpp", - "hash_map": "cpp", - "slist": "cpp", - "valarray": "cpp", - "bufferobject": "cpp", - "image": "cpp", - "texture": "cpp", - "buffered_value": "cpp" - } -} \ No newline at end of file diff --git a/CMakeLists.txt b/CMakeLists.txt index ddc2485..773a24b 100644 --- a/CMakeLists.txt +++ b/CMakeLists.txt @@ -2,9 +2,9 @@ cmake_minimum_required(VERSION 3.10) project(BundleTrack) -set(CMAKE_BUILD_TYPE Release) +set(CMAKE_BUILD_TYPE Debug) add_compile_options(-std=c++14 -fpermissive -fPIC -g) -set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xcompiler;-fPIC;-gencode arch=compute_30,code=sm_30;-gencode arch=compute_61,code=sm_61;-gencode arch=compute_75,code=sm_75;-O3;-std=c++11;-use_fast_math;--default-stream per-thread) +set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xcompiler;-fPIC;-gencode arch=compute_86,code=sm_86;-O3;-std=c++11;-use_fast_math;--default-stream per-thread) if (${CMAKE_BUILD_TYPE} STREQUAL "Release") set(COMPILE_OPTIONS -fopenmp -march=native -msse -msse2 -msse3 -msse4 -msse4.2 -mavx2 -ftree-vectorize) diff --git a/README.md b/README.md index 887fca9..a3c9a3f 100644 --- a/README.md +++ b/README.md @@ -1,194 +1,9 @@ -This is the official implementation of our paper: +# BundleTrack2 -[BundleTrack: 6D Pose Tracking for Novel Objects without Instance or Category-Level 3D Models](https://arxiv.org/abs/2108.00516) +I want to modify BudleTrack, to make it compile on the latest machine. - accepted in International Conference on Intelligent Robots and Systems (IROS) 2021. +# Build BundleTrack from scratch -# Abstract -Most prior 6D object pose tracking often assume that the target object's CAD model, at least at a category-level, is available for offline training or during online template matching. This work proposes BundleTrack, a general framework for 6D pose tracking of novel objects, which does not depend upon 3D models, either at the instance or category-level. It leverages the complementary attributes of recent advances in deep learning for segmentation and robust feature extraction, as well as memory-augmented pose graph optimization for spatiotemporal consistency. This enables long-term, low-drift tracking under various challenging scenarios, including significant occlusions and object motions. Comprehensive experiments given two public benchmarks demonstrate that the proposed approach significantly outperforms state-of-art, category-level 6D tracking or dynamic SLAM methods. When compared against state-of-art methods that rely on an object instance CAD model, comparable performance is achieved, despite the proposed method's reduced information requirements. An efficient implementation in CUDA provides a real-time performance of 10Hz for the entire framework. - -

- - -

- - - -**This repo can be readily applied to 6D pose tracking for novel unknown objects. For CAD model-based 6D pose tracking, please check out my another repository of [se(3)-TrackNet](https://github.com/wenbowen123/iros20-6d-pose-tracking)** - - -# Bibtex -```bibtex -@inproceedings{wen2021bundletrack, - title={BundleTrack: 6D Pose Tracking for Novel Objects without Instance or Category-Level 3D Models}, - author={Wen, B and Bekris, Kostas E}, - booktitle={IEEE/RSJ International Conference on Intelligent Robots and Systems}, - year={2021} -} -``` - - - -# Supplementary Video -Click to watch - -[](https://www.youtube.com/watch?v=89pnv3M_84g) - - - -# IROS 2021 Presentation -Click to watch - -[](https://www.youtube.com/watch?v=0UorLR0ADd4) - - -# Results - - - - - -# Benchmark Output Results -For convenience of benchmarking and making plots, results of pose outputs can be downloaded below -- For NOCS Dataset: https://archive.cs.rutgers.edu/archive/a/2021/pracsys/2021_iros_bundletrack/nocs_ours_results_iros2021.tar.gz -- For YCBInEOAT: https://archive.cs.rutgers.edu/archive/a/2021/pracsys/2021_iros_bundletrack/ycbineoat_ours_results.tar.gz - - - -# Setup -For the environment setup, it's strongly recommended to use our provided docker environment (setting up from scratch is very complicated and not supported in this repo). For this, you don't have to know how docker works. Only some basic commands are needed and will be provided in the below steps. - -- Install docker (https://docs.docker.com/get-docker/). - -- Run - ``` - docker pull wenbowen123/bundletrack:latest - docker pull wenbowen123/lf-net-release-env:latest - ``` - -- Edit the docker/run_container.sh, update the paths of `BUNDLETRACK_DIR`, `NOCS_DIR` and `YCBINEOAT_DIR` - -- Run `bash docker/run_container.sh` - -- `cd [PATH_TO_BUNDLETRACK]` - -- `rm -rf build && mkdir build && cd build && cmake .. && make` - - -# Data -Depending on what you want to run, download those data that are neccessary. -- [Download weights of feature detection network](https://archive.cs.rutgers.edu/archive/a/2021/pracsys/2021_iros_bundletrack/indoor.tar.gz), extract and put it under `lf-net-release/release/models`, so it'll be `BundleTrack/lf-net-release/release/models/indoor` -- [Download weights of video segmentation network](https://archive.cs.rutgers.edu/archive/a/2021/pracsys/2021_iros_bundletrack/pretrained.tar.gz), extract and put it under `./transductive-vos.pytorch`, so it'll be `BundleTrack/transductive-vos.pytorch/pretrained` -- [Download our precomputed masks](https://archive.cs.rutgers.edu/archive/a/2021/pracsys/2021_iros_bundletrack/masks.tar.gz), extract and put in the repo so it becomes `BundleTrack/masks` - -- [Download NOCS Dataset](https://github.com/hughw19/NOCS_CVPR2019), then put it under a folder named "NOCS". [Download the converted ground-truth text pose files](https://archive.cs.rutgers.edu/archive/a/2021/pracsys/2021_iros_bundletrack/real_test_text.tar.gz). And [download the addon](https://drive.google.com/file/d/1BknMsoRKRV-nhLDpozog2Lrz1OWTKqkM/view?usp=sharing). Finally unzip the files and make sure the path structure is like this: - - ``` - NOCS - ├── NOCS-REAL275-additional - ├── real_test - ├── gts - | └── real_test_text - └── obj_models - ``` - -- [Download YCBInEOAT](https://archive.cs.rutgers.edu/archive/a/2020/pracsys/Bowen/iros2020/YCBInEOAT/) to make it like - ``` - YCBInEOAT - ├── bleach0 - | ├──annotated_poses - | ├──depth - | ├──depth_filled - | ├──gt_mask - | ├──masks - | ├──masks_vis - | ├──rgb - | ├──cam_K.txt - | ├──...... - └── .... - ``` - -- [Download YCB Objects](http://ycb-benchmarks.s3-website-us-east-1.amazonaws.com/) - -# Run predictions on NOCS -- Open a separate terminal and run - - ``` - bash lf-net-release/docker/run_container.sh - cd [PATH_TO_BUNDLETRACK] - cd lf-net-release && python run_server.py - ``` - - -- Go back to the terminal where you launched the bundletrack docker in above and run below. The output will be saved to `debug_dir` specified in config file. By default it's `/tmp/BundleTrack/`. For more detailed logs, change `LOG` to 2 or higher in `config_nocs.yml`. - - ``` - python scripts/run_nocs.py --nocs_dir [PATH_TO_NOCS] --scene_id 1 --port 5555 --model_name can_arizona_tea_norm - ``` - -- Finally, the results will be saved in `/tmp/BundleTrack/` - - -- For evaluating on the entire NOCS Dataset, run (**NOTE that this will add noise to perturb the initial ground-truth pose for evaluation as explained in the paper. If you want to see how BundleTrack actually performs, run the above section**) - ``` - python scripts/eval_nocs.py --nocs_dir [PAHT TO NOCS] --results_dir [PATH TO THE RUNNING OUTPUTS] - ``` - - -# Run predictions on YCBInEOAT - -- Change the `model_name` and `model_dir` in `config_ycbineoat.yml` to the path to the .obj file (e.g. For folder `bleach0`, the model_name is `021_bleach_cleanser`, and model_dir is `[Your path to YCB Objects]/021_bleach_cleanser/textured_simple.obj`) - -- Open a separate terminal and run - - ``` - bash lf-net-release/docker/run_container.sh - cd [PATH_TO_BUNDLETRACK] - cd lf-net-release && python run_server.py - ``` - -- Go back to the terminal where you launched the bundletrack docker in above, and run below. The output will be saved to `debug_dir` specified in config file. By default it's `/tmp/BundleTrack/` - - ``` - python scripts/run_ycbineoat.py --data_dir [PATH_TO_YCBInEOAT] --port 5555 --model_name [The YCB object's name, e.g. 021_bleach_cleanser] - ``` - -- Finally, the results will be saved in `/tmp/BundleTrack/`. For more detailed logs, change `LOG` to 2 or higher in `config_ycbineoat.yml`. - -- For evaluating on the entire YCBInEOAT Dataset, run - ``` - python scripts/eval_nocs.py --ycbineoat_dir [PAHT TO YCBINEOAT] --ycb_model_dir [YCB MODELS FOLDER] --results_dir [PATH TO THE RUN OUTPUTS] - ``` - - -# Run predictions on your own RGBD data - -- Download YCBInEOAT, if you haven't done so in above. - -- Open a separate terminal and run - - ``` - bash lf-net-release/docker/run_container.sh - cd [PATH_TO_BUNDLETRACK] - cd lf-net-release && python run_server.py - ``` - -- Prepare segmentation masks. In YCBInEOAT Dataset, we computed masks from robotic arm forward kinematics. If your scene is not too complicated similar to NOCS Dataset, you can run the video segmentation network to get masks as below: - - - First you need to prepare an initial mask (grayscale image, where 0 means background, else foreground). - - `python transductive-vos.pytorch/run_video.py --img_dir [THE PATH TO COLOR FILES] --init_mask_file [THE INITIAL MASK FILE YOU PREPARED ABOVE] --mask_save_dir [WHERE TO SAVE]` - - Prepare your folder structure same as any folder (e.g. "mustard_easy_00_02") in YCBInEOAT Dataset. Put it under the same directory in YCBInEOAT, i.e. next to "mustard_easy_00_02". Then edit `config_ycbineoat.yml` to make sure the paths at top are right. - - Structure: - ``` - mustard_easy_00_02 - ├── rgb - ├── masks - ├── depth - └── cam_K.txt - ``` - -- Go back to the terminal where you launched the bundletrack docker, run below. The output will be saved to `debug_dir` specified in config file. By default it's `/tmp/BundleTrack/` - ``` - python scripts/run_ycbineoat.py --data_dir [PATH TO YOUR FOLDER ABOVE] --port 5555 - ``` +1. Install LibZMQ. +2. Install opencv with cuda support. +3. Cooperate R2D2 with BundleTrack. \ No newline at end of file diff --git a/config_ycbineoat.yml b/config_ycbineoat.yml index 6791825..14b1a4b 100644 --- a/config_ycbineoat.yml +++ b/config_ycbineoat.yml @@ -1,9 +1,9 @@ -data_dir: /media/bowen/e25c9489-2f57-42dd-b076-021c59369fec/catkin_ws/src/iros20_dataset/video_rosbag/IROS_SELECTED/FINISHED_LABEL.iros_submission_version/bleach0 -mask_dir: '/media/bowen/e25c9489-2f57-42dd-b076-021c59369fec/catkin_ws/src/iros20_dataset/video_rosbag/IROS_SELECTED/FINISHED_LABEL.iros_submission_version/bleach0/masks' -model_name: 021_bleach_cleanser -model_dir: /media/bowen/e25c9489-2f57-42dd-b076-021c59369fec/DATASET/YCB_Video_Dataset/CADmodels/021_bleach_cleanser/textured_simple.obj +data_dir: /home/shreesh/DATASET/YCBInEOAT/cracker_box_icg_1 +mask_dir: '/home/shreesh/DATASET/YCBInEOAT/cracker_box_icg_1/masks' +model_name: 003_cracker_box +model_dir: /home/shreesh/DATASET/YCBInEOAT/ycb/003_cracker_box/google_512k/textured.obj debug_dir: /tmp/BundleTrack/ -LOG: 0 +LOG: 2 port: '5555' depth_processing: diff --git a/docker/run_container.sh b/docker/run_container.sh index 63cb279..3a14f66 100644 --- a/docker/run_container.sh +++ b/docker/run_container.sh @@ -1,6 +1,6 @@ -BUNDLETRACK_DIR="/home/bowen/debug/BundleTrack" -NOCS_DIR="/media/bowen/e25c9489-2f57-42dd-b076-021c59369fec/DATASET/NOCS" -YCBINEOAT_DIR="/iros_submission_version" +BUNDLETRACK_DIR="/home/shreesh/BundleTrack" +NOCS_DIR="/home/shreesh/DATASET/NOCS" +YCBINEOAT_DIR="/home/shreesh/DATASET/YCBInEOAT" echo "BUNDLETRACK_DIR $BUNDLETRACK_DIR" echo "NOCS_DIR $NOCS_DIR" echo "YCBINEOAT_DIR $YCBINEOAT_DIR" diff --git a/fnn/run_server.py b/fnn/run_server.py new file mode 100644 index 0000000..70472e0 --- /dev/null +++ b/fnn/run_server.py @@ -0,0 +1,62 @@ +#!/usr/bin/env python +from __future__ import print_function +import numpy as np +import zmq +import cv2 +from tqdm import tqdm + + +def empty_network(feed_dict): + empty_output = { + "kpts": np.zeros((0, 2), dtype=np.float32), + "feats": np.zeros((0, 128), dtype=np.float32), + } + return empty_output + + +if __name__ == "__main__": + import argparse + parser = argparse.ArgumentParser() + parser.add_argument("--config", type=str, default="configs/r2d2.yml") + parser.add_argument("--ckpt", type=str, default="checkpoints/r2d2/r2d2.ckpt") + parser.add_argument("--port", type=int, default=5555) + args = parser.parse_args() + + context = zmq.Context() + socket = context.socket(zmq.REP) + port = f"tcp://*:{args.port}" + print("port", port) + socket.bind(port) + + while 1: + print(f"FNN listending to {port}") + msgs = socket.recv_multipart(0) + assert len(msgs) == 2, "#msgs={}".format(len(msgs)) + wh = np.frombuffer(msgs[0], dtype=np.int32) + W = wh[0] + H = wh[1] + print(f"W={W}, H={H}") + msg = msgs[1] + photo = np.frombuffer(msg, dtype=np.uint8).reshape(H, W, -1).squeeze() + photo_ori = photo.copy() + + rgb = photo.copy() + if photo.ndim == 3 and photo.shape[-1] == 3: + photo = cv2.cvtColor(photo, cv2.COLOR_RGB2GRAY) + photo = photo[None, ..., None].astype(np.float32) / 255.0 + assert photo.ndim == 4 + + feed_dict = { + "photo_ph": photo, + } + + outs = empty_network(feed_dict=feed_dict) + + num_feat = len(outs["kpts"]) + feat_dim = outs["feats"].shape[1] + msg = np.array([num_feat, feat_dim]).reshape(-1).astype(np.int32).tobytes() + socket.send(msg, 2) + msg = outs["kpts"].astype(np.float32).reshape(-1).tobytes() + socket.send(msg, 2) + msg = outs["feats"].astype(np.float32).reshape(-1).tobytes() + socket.send(msg, 0) diff --git a/r2d2 b/r2d2 new file mode 160000 index 0000000..251ffed --- /dev/null +++ b/r2d2 @@ -0,0 +1 @@ +Subproject commit 251ffedd5ddce530560b44da4b5645e83db4a20d diff --git a/scripts/run_ycbineoat.py b/scripts/run_ycbineoat.py index 22d7375..82eb98e 100644 --- a/scripts/run_ycbineoat.py +++ b/scripts/run_ycbineoat.py @@ -46,18 +46,19 @@ pass -def run_one_video(data_dir,model_name,model_dir,cfg1,port): +def run_one_video(data_dir,model_name,model_dir,custom_yc,cfg1,port): cfg = copy.deepcopy(cfg1) - name = data_dir.split('/')[-2] + name = data_dir.split('/')[-1] cur_out_dir = '/tmp/BundleTrack/ycbineoat/{}/'.format(name) os.system(f'mkdir -p {cur_out_dir}') cfg['data_dir'] = data_dir - cfg['mask_dir'] = f'{code_dir}/masks/ycbineoat/{name}/masks' + cfg['mask_dir'] = f'/home/shreesh/DATASET/YCBInEOAT/{name}/masks' cfg['model_name'] = model_name cfg['model_dir'] = model_dir cfg['debug_dir'] = cur_out_dir + cfg['custom_yc'] = custom_yc cfg['LOG'] = 0 cfg['port'] = port tmp_config_dir = '/tmp/config_{}.yml'.format(name) @@ -76,10 +77,11 @@ def run_one_video(data_dir,model_name,model_dir,cfg1,port): if __name__=='__main__': parser = argparse.ArgumentParser() - parser.add_argument('--data_dir', type=str, default='/media/bowen/e25c9489-2f57-42dd-b076-021c59369fec/catkin_ws/src/iros20_dataset/video_rosbag/IROS_SELECTED/FINISHED_LABEL.iros_submission_version/bleach0') + parser.add_argument('--data_dir', type=str, default='/home/shreesh/DATASET/YCBInEOAT/cracker_box_icg_1') parser.add_argument('--port', type=int, default=5555) - parser.add_argument('--model_name', type=str, default='021_bleach_cleanser') - parser.add_argument('--model_dir', type=str, default='/media/bowen/e25c9489-2f57-42dd-b076-021c59369fec/DATASET/YCB_Video_Dataset/CADmodels/021_bleach_cleanser/textured.obj') + parser.add_argument('--model_name', type=str, default='003_cracker_box') + parser.add_argument('--model_dir', type=str, default='/home/shreesh/DATASET/YCBInEOAT/ycb/003_cracker_box/google_512k/textured.obj') + parser.add_argument('--custom_yc', type=str, default='false') args = parser.parse_args() @@ -89,8 +91,8 @@ def run_one_video(data_dir,model_name,model_dir,cfg1,port): raise RuntimeError(f"Make sure data_dir={data_dir} exists") code_dir = os.path.dirname(os.path.realpath(__file__)) - config_dir = f'{code_dir}/../config_ycbineoat.yml' + config_dir = '/home/shreesh/BundleTrack/config_ycbineoat.yml' with open(config_dir,'r') as ff: cfg = yaml.safe_load(ff) - run_one_video(args.data_dir,args.model_name,args.model_dir,cfg,args.port) + run_one_video(args.data_dir,args.model_name,args.model_dir,args.custom_yc,cfg,args.port) diff --git a/src/Bundler.cpp b/src/Bundler.cpp index 68d675f..53e8220 100644 --- a/src/Bundler.cpp +++ b/src/Bundler.cpp @@ -160,6 +160,10 @@ void Bundler::processNewFrame(std::shared_ptr frame) if (frame->_id==0) { + std::vector initialBBoxCenter = Utils::getBoundingBoxInitialCenterModelFree(frame->_depth,frame->_fg_mask,_data_loader->_K); + frame->_x_bbox_init = initialBBoxCenter[0]; + frame->_y_bbox_init = initialBBoxCenter[1]; + frame->_z_bbox_init = initialBBoxCenter[2]; checkAndAddKeyframe(frame); return; } @@ -395,12 +399,28 @@ void Bundler::saveNewframeResult() } } } - Utils::drawProjectPoints(cur_model,_data_loader->_K,color_viz); + + if (_newframe->_id==0) + { + std::vector initialBBoxCenter = Utils::getBoundingBoxInitialCenterModelFree(_newframe->_depth,_newframe->_fg_mask,_data_loader->_K); + _newframe->_x_bbox_init = initialBBoxCenter[0]; + _newframe->_y_bbox_init = initialBBoxCenter[1]; + _newframe->_z_bbox_init = initialBBoxCenter[2]; + } + + Utils::drawProjectPoints(cur_model,_data_loader->_K,color_viz,_newframe->_pose_in_model); + Utils::drawBBoxPCL(_newframe->_x_bbox_init,_newframe->_y_bbox_init,_newframe->_z_bbox_init,_data_loader->_K,color_viz,_newframe->_pose_in_model); cv::putText(color_viz,_newframe->_id_str,{5,30},cv::FONT_HERSHEY_PLAIN,2,{255,0,0},1,8,false); // cv::imshow("color_viz",color_viz); // cv::waitKey(1); - cv::imwrite(debug_dir+"/color_viz/"+_newframe->_id_str+"_color_viz.jpg",color_viz,{CV_IMWRITE_JPEG_QUALITY, 80}); - cv::imwrite(out_dir+"color_viz.jpg",color_viz,{CV_IMWRITE_JPEG_QUALITY, 80}); + cv::imwrite(debug_dir+"/color_viz/"+_newframe->_id_str+"_color_viz.jpg",color_viz,{cv::IMWRITE_JPEG_QUALITY, 80}); + const std::string color_viz_out_dir = debug_dir+"color_viz/"; + if (!boost::filesystem::exists(color_viz_out_dir)) + { + system(std::string("mkdir -p "+color_viz_out_dir).c_str()); + } + + cv::imwrite(color_viz_out_dir+_newframe->_id_str+"_color_viz.jpg",color_viz,{CV_IMWRITE_JPEG_QUALITY, 80}); const std::string raw_dir = debug_dir+"/color_raw/"; if (!boost::filesystem::exists(raw_dir)) diff --git a/src/DataLoader.cpp b/src/DataLoader.cpp index 1135426..9445282 100644 --- a/src/DataLoader.cpp +++ b/src/DataLoader.cpp @@ -383,4 +383,45 @@ std::shared_ptr DataLoaderYcbineoat::next() return frame; } +std::shared_ptr DataLoaderYcbineoat::nextCustom() +{ + assert(_id<_color_files.size()); + const std::string data_dir = (*yml)["data_dir"].as(); + + std::string color_file = _color_files[_id]; + std::cout<<"color file: "< strs; + boost::split(strs, color_file, boost::is_any_of("/")); + boost::split(strs, strs.back(), boost::is_any_of(".")); + index_str = strs[0]; + } + + cv::Mat depth_raw; + std::string depth_dir = data_dir+"/depth/"+index_str+".depth.png"; + Utils::readDepthImage(depth_raw, depth_dir); + + cv::Mat depth_sim; + depth_sim = depth_raw.clone(); + + cv::Mat depth; + depth = depth_raw.clone(); + + Eigen::Matrix4f pose(Eigen::Matrix4f::Identity()); + if (_id==0) + { + pose = _ob_in_cam0.inverse(); + } + + Eigen::Vector4f roi; + roi << 99999,0,99999,0; + + std::shared_ptr frame(new Frame(color,depth,depth_raw,depth_sim, roi, pose, _id, index_str+".seg", _K, yml, NULL, _real_model)); + _id++; + + return frame; +} + diff --git a/src/DataLoader.h b/src/DataLoader.h index 7869a21..a8ffda7 100644 --- a/src/DataLoader.h +++ b/src/DataLoader.h @@ -87,7 +87,7 @@ class DataLoaderYcbineoat : public DataLoaderBase DataLoaderYcbineoat(std::shared_ptr yml1); ~DataLoaderYcbineoat(); std::shared_ptr next(); - + std::shared_ptr nextCustom(); }; diff --git a/src/FeatureManager.cpp b/src/FeatureManager.cpp index dbdde17..a335ee4 100644 --- a/src/FeatureManager.cpp +++ b/src/FeatureManager.cpp @@ -760,7 +760,12 @@ int SiftManager::countInlierCorres(std::shared_ptr frameA, std::shared_pt void SiftManager::vizCorresBetween(std::shared_ptr frameA, std::shared_ptr frameB, const std::string &name) { if ((*yml)["LOG"].as()<1) return; - const std::string out_dir = (*yml)["debug_dir"].as()+"/"+_bundler->_newframe->_id_str+"/"; + const std::string out_dir = (*yml)["debug_dir"].as()+"/match_viz/"; + // std::cout<_id_str+"_match_"+frameB->_id_str+"_"+name+".jpg"; std::ifstream tmp(out_match_file); cv::Mat colorA = frameA->_color.clone(); @@ -791,7 +796,7 @@ void SiftManager::vizCorresBetween(std::shared_ptr frameA, std::shared_pt cv::Mat out; cv::drawMatches( colorA, kptsA, colorB, kptsB, cv_matches, out, cv::Scalar::all(-1), cv::Scalar::all(-1), std::vector(), cv::DrawMatchesFlags::NOT_DRAW_SINGLE_POINTS ); - cv::imwrite(out_match_file, out, {CV_IMWRITE_JPEG_QUALITY,80}); + cv::imwrite(out_match_file, out, {cv::IMWRITE_JPEG_QUALITY,80}); } diff --git a/src/Frame.cpp b/src/Frame.cpp index 3ea1ce8..23c2c1e 100644 --- a/src/Frame.cpp +++ b/src/Frame.cpp @@ -63,6 +63,9 @@ Frame::Frame(const cv::Mat &color, const cv::Mat &depth, const cv::Mat &depth_ra _real_model = real_model; _pose_inited = false; _roi = roi; + _x_bbox_init = 0; + _y_bbox_init = 0; + _z_bbox_init = 0; const int n_pixels = _H*_W; cudaMalloc(&_depth_gpu, n_pixels*sizeof(float)); @@ -70,7 +73,7 @@ Frame::Frame(const cv::Mat &color, const cv::Mat &depth, const cv::Mat &depth_ra cudaMalloc(&_color_gpu, n_pixels*sizeof(uchar4)); - cv::cvtColor(_color, _gray, CV_BGR2GRAY); + cv::cvtColor(_color, _gray, cv::COLOR_BGR2GRAY); updateDepthGPU(); processDepth(); diff --git a/src/Frame.h b/src/Frame.h index 1663ab2..c26302f 100644 --- a/src/Frame.h +++ b/src/Frame.h @@ -69,6 +69,7 @@ class Frame Status _status; bool _pose_inited; std::map, std::shared_ptr> _map_points; + float _x_bbox_init,_y_bbox_init,_z_bbox_init; float *_depth_gpu; uchar4 *_color_gpu; diff --git a/src/Utils.cpp b/src/Utils.cpp index 2b31dca..ae2e4ef 100644 --- a/src/Utils.cpp +++ b/src/Utils.cpp @@ -111,6 +111,39 @@ void convert3dOrganizedRGB(cv::Mat &objDepth, cv::Mat &colImage, Eigen::Matrix3f } } } + + +std::vector getBoundingBoxInitialCenterModelFree(cv::Mat &objDepth, cv::Mat &fgMask, Eigen::Matrix3f &camIntrinsic) +{ + const int imgWidth = objDepth.cols; + const int imgHeight = objDepth.rows; + + float x_mean = 0; + float y_mean = 0; + float z_mean = 0; + int num_points = 0; + + for(int u = 0; u < imgHeight; u++) + { + for(int v = 0; v < imgWidth; v++) + { + float depth = objDepth.at(u,v); + if(fgMask.at(u,v) == 0 || depth <= 0.1 || depth >= 2.0) + continue; + x_mean += (float)((v - camIntrinsic(0,2)) * depth / camIntrinsic(0,0)); + y_mean += (float)((u - camIntrinsic(1,2)) * depth / camIntrinsic(1,1)); + z_mean += depth; + } + } + + x_mean /= (float)num_points; + y_mean /= (float)num_points; + z_mean /= (float)num_points; + + int gt_mean[3] = {x_mean,y_mean,z_mean}; + return std::vector(gt_mean,gt_mean+3); +} + template void convert3dOrganizedRGB(cv::Mat &objDepth, cv::Mat &colImage, Eigen::Matrix3f &camIntrinsic, boost::shared_ptr> objCloud); template void convert3dOrganizedRGB(cv::Mat &objDepth, cv::Mat &colImage, Eigen::Matrix3f &camIntrinsic, boost::shared_ptr> objCloud); @@ -218,7 +251,31 @@ void solveRigidTransformBetweenPoints(const Eigen::MatrixXf &points1, const Eige } -void drawProjectPoints(PointCloudRGBNormal::Ptr cloud, const Eigen::Matrix3f &K, cv::Mat &out) +std::pair getProjectPointCoord3Dto2D(float x, float y, float z, const Eigen::Matrix3f &K, cv::Mat &out) { + + int u = std::round(x*K(0,0)/z + K(0,2)); + int v = std::round(y*K(1,1)/z + K(1,2)); + if(u < 0) + u = 0; + else if(u >= out.cols) + u = (out.cols - 1); + if(v < 0) + v = 0; + else if(v >= out.rows) + v = (out.rows - 1); + + cv::circle(out, {u,v}, 1, {0,255,255}, -1); + + return std::pair(u,v); +} + +void drawLine(cv::Mat &out, int x1, int y1, int x2, int y2, cv::Scalar color) { + cv::Point p1(x1,y1); + cv::Point p2(x2,y2); + cv::line(out,p1,p2,color,2); +} + +void drawProjectPoints(PointCloudRGBNormal::Ptr cloud, const Eigen::Matrix3f &K, cv::Mat &out, const Eigen::Matrix4f &pose) { Utils::downsamplePointCloud(cloud,cloud,0.01); for (const auto &pt:cloud->points) @@ -230,6 +287,101 @@ void drawProjectPoints(PointCloudRGBNormal::Ptr cloud, const Eigen::Matrix3f &K, } } +std::vector getMeanCoordinatesFromPCLCloud(PointCloudRGBNormal::Ptr initialCloud) +{ + float x_mean = 0; + float y_mean = 0; + float z_mean = 0; + float x_sum = 0; + float y_sum = 0; + float z_sum = 0; + int num_points = 0; + for (const auto &pt:initialCloud->points) + { + x_sum += pt.x; + y_sum += pt.y; + z_sum += pt.z; + num_points++; + } + + x_mean = x_sum/num_points; + y_mean = y_sum/num_points; + z_mean = z_sum/num_points; + + float gt_mean[] = {x_mean,y_mean,z_mean}; + return std::vector(gt_mean,gt_mean+3); +} + +void drawBBoxPCL(float x, float y, float z, const Eigen::Matrix3f &K, cv::Mat &out, const Eigen::Matrix4f &pose) +{ + + pcl::PointCloud bbox; + float x_adds[2] = {-0.05,0.05}; + float y_adds[2] = {-0.05,0.05}; + float z_adds[2] = {-0.05,0.05}; + for(auto y_add : y_adds) + for(auto x_add : x_adds) + for(auto z_add : z_adds) + bbox.push_back(pcl::PointXYZ(x + x_add,y + y_add,z + z_add)); + + pcl::PointCloud bboxTrans; + pcl::transformPointCloud(bbox,bboxTrans,pose.inverse()); + + std::pair p[8]; + int idx = 0; + for(const auto &pt : bboxTrans.points) + { + p[idx] = getProjectPointCoord3Dto2D(pt.x,pt.y,pt.z,K,out); + idx++; + } + + int lineStartingPoints[12] = {0,0,0,1,1,2,2,3,4,4,5,6}; + int lineEndingPoints[12] = {1,2,4,3,5,3,6,7,5,6,7,7}; + + for(int i = 0; i < 12; i++) + drawLine(out,p[lineStartingPoints[i]].first,p[lineStartingPoints[i]].second,p[lineEndingPoints[i]].first,p[lineEndingPoints[i]].second,cv::Scalar(0,255,0)); +} + +void drawBBox(float x, float y, float z, const Eigen::Matrix3f &K, cv::Mat &out, const Eigen::Matrix4f &pose) +{ + // pcl::PointCloud bbox; + Eigen::MatrixXf bbox = Eigen::MatrixXf::Random(4,9); + int pointIdx = 0; + float x_adds[2] = {-0.05,0.05}; + float y_adds[2] = {-0.05,0.05}; + float z_adds[2] = {-0.05,0.05}; + for(auto y_add : y_adds) + for(auto x_add : x_adds) + for(auto z_add : z_adds) + { + bbox(0,pointIdx) = x + x_add; + bbox(1,pointIdx) = y + y_add; + bbox(2,pointIdx) = z + z_add; + bbox(3,pointIdx) = 1; + pointIdx++; + } + + bbox(0,8) = x; + bbox(1,8) = y; + bbox(2,8) = z; + + // pcl::PointCloud bboxTrans; + // pcl::transformPointCloud(bbox,bboxTrans,pose.inverse()); + Eigen::MatrixXf bboxTrans = pose*bbox; + + std::pair p[9]; + for(int idx = 0; idx < 9; idx++) + p[idx] = getProjectPointCoord3Dto2D(bboxTrans(0,idx),bboxTrans(1,idx),bboxTrans(2,idx),K,out); + + cv::circle(out, {p[8].first,p[8].second}, 4, {0,0,255}, -1); + + int lineStartingPoints[12] = {0,0,0,1,1,2,2,3,4,4,5,6}; + int lineEndingPoints[12] = {1,2,4,3,5,3,6,7,5,6,7,7}; + + for(int i = 0; i < 12; i++) + drawLine(out,p[lineStartingPoints[i]].first,p[lineStartingPoints[i]].second,p[lineEndingPoints[i]].first,p[lineEndingPoints[i]].second,cv::Scalar(0,255,0)); +} + } // namespace Utils diff --git a/src/Utils.h b/src/Utils.h index 1e0f8f5..79a62b1 100755 --- a/src/Utils.h +++ b/src/Utils.h @@ -179,7 +179,11 @@ bool isPixelInsideImage(const int H, const int W, float u, float v); void solveRigidTransformBetweenPoints(const Eigen::MatrixXf &points1, const Eigen::MatrixXf &points2, Eigen::Matrix4f &pose); -void drawProjectPoints(PointCloudRGBNormal::Ptr cloud, const Eigen::Matrix3f &K, cv::Mat &out); +void drawProjectPoints(PointCloudRGBNormal::Ptr cloud, const Eigen::Matrix3f &K, cv::Mat &out, const Eigen::Matrix4f &pose); +std::vector getMeanCoordinatesFromPCLCloud(PointCloudRGBNormal::Ptr initialCloud); +std::vector getBoundingBoxInitialCenterModelFree(cv::Mat &objDepth, cv::Mat &fgMask, Eigen::Matrix3f &camIntrinsic); +void drawBBoxPCL(float x, float y, float z, const Eigen::Matrix3f &K, cv::Mat &out, const Eigen::Matrix4f &pose); +void drawBBox(float x, float y, float z, const Eigen::Matrix3f &K, cv::Mat &out, const Eigen::Matrix4f &pose); template void parseMatrixTxt(std::string filename, Eigen::Matrix &out) diff --git a/src/app/bundle_track_ycbineoat.cpp b/src/app/bundle_track_ycbineoat.cpp index cf96249..8e240de 100644 --- a/src/app/bundle_track_ycbineoat.cpp +++ b/src/app/bundle_track_ycbineoat.cpp @@ -63,12 +63,15 @@ int main(int argc, char **argv) Bundler bundler(yml,&data_loader); + const bool custom_yc = (*yml)["custom_yc"].as(); + while (data_loader.hasNext()) { - std::shared_ptr frame = data_loader.next(); + std::shared_ptr frame = (custom_yc) ? data_loader.nextCustom() : data_loader.next(); if (!frame) break; const std::string index_str = frame->_id_str; - const std::string out_dir = (*yml)["debug_dir"].as()+"/"+index_str+"/"; + const std::string out_dir = (*yml)["debug_dir"].as(); + cv::imwrite(out_dir+index_str+"_color.png",frame->_color); Eigen::Matrix4f cur_in_model(data_loader._ob_in_cam0.inverse());