tym-b / implicit-slam Goto Github PK
View Code? Open in Web Editor NEWDense SLAM with an Implicit Neural Representation
Dense SLAM with an Implicit Neural Representation
Unable to run in Colab. The error happened at
mappingThread = MappingThread(sharedData)
ValueError: tf.function only supports singleton tf. Variables created on the first cal1. Make sure the tf.Variable is onlycreated once or created outside tf.function.See https:/ww.tensorflow.org/guide/function#creating_tfvariables for more information.
I am currently trying to replicate the results from the paper on the Replica scenes . I am working with the room-0 in which I have recorded a random path of 2k frames using the Habitat simulator. I have randomly selected locations in the scene and linked those locations using shortest path (see green path below). I have an agent step size of 1cm forward and 1.0deg rotations. And I render one RGB-D frame at each time step.
I have a couple of questions:
(1) For the experiments using the Replica dataset, do you know how the random trajectories were generated? Was it by hand, if so, using which software? Or automatically (using which algorithm?)? On the figure 5 of the paper, I see a nice curvy random trajectory and I would like to replicate this data. The paper says it rendered 2000 frames from a randomly generated path. If you know about the path generation that would be great if you could you elaborate on this - like with details on the frame rate, agent step sizes (if any?), etc..
(2) What hyper parameters should be used for the Replica experiments? Currently I am using the following:
uniform_samples_per_region = 4
tracking_lr = 0.1
tracking_iterations_per_frame = 10
n_rays_per_tracking_iter = 240
registration_threshold tp = 0.65
photo-metric loss weight = 5
window size = 5
pixel samples = 200
I get the following result on an A40 GPU:
#mapping iterations: 8,355
#tracking iterations: 19,656
depth residual: 0.12
RGB residual: 0.06
The tracking is just Ok. It seems far less accurate than what you have on Fig.5. In addition, the reconstruction seems very poor. Do you have any suggestions on how I could improve this result?
Hello, tymoteuszb,
Thanks for your great work of IMAP implement. However, I meet the problem when i running the code below
physical_devices = tf.config.list_physical_devices('GPU')
tf.config.set_logical_device_configuration(physical_devices[0],
[tf.config.LogicalDeviceConfiguration(memory_limit=6144, experimental_priority=-1),
tf.config.LogicalDeviceConfiguration(memory_limit=6144, experimental_priority=0)])
tf.config.list_logical_devices()
Report that
Virtual devices cannot be modified after being initialized.
I wonder if you meet the same problem before. It's soooooo confusing!
Bset wish, James·GZL
W tensorflow/core/framework/op_kernel.cc:1745] OP_REQUIRES failed at xla_ops.cc:248 : INTERNAL: Non-root tuple types are not handled.
Traceback (most recent call last):
File "D:/cameraLocalization/imaptf/complete.py", line 859, in
mappingThread = MappingThread(sharedData)
File "D:/cameraLocalization/imaptf/complete.py", line 693, in init
self.optimize()
File "D:/cameraLocalization/imaptf/complete.py", line 712, in optimize
self.poses_optimizer)
File "D:\Anaconda\envs\imap\lib\site-packages\tensorflow\python\util\traceback_utils.py", line 153, in error_handler
raise e.with_traceback(filtered_tb) from None
File "D:\Anaconda\envs\imap\lib\site-packages\tensorflow\python\eager\execute.py", line 59, in quick_execute
inputs, attrs, num_outputs)
tensorflow.python.framework.errors_impl.InternalError: Non-root tuple types are not handled. [Op:__inference_bundle_adjust_283900]
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.