nvlabs / flat Goto Github PK

Hello,

When I try the init.py file, the files downloaded by googledrivedownloader are all not correct with only 404 information inside.
I try to use gdown but still can not get the file.
I check the issues and found that the google drive link mentioned in #11 is no longer valid.

I am not sure if the problem occurs due to my operation or if the google drive id in init.py is not usable now.
But I would appreciate it if you could help download the dataset.

Thanks and best regards.

Hi,
I'm interested in using the FLAT dataset but I have some questions regarding the scenes used for training and testing.
I started by downloading the kinect data:
python init.py -c kinect
Which seems to download ~2000 measurements w/ ground truth depth.

However, when I check the train.txt file, only one example is listed there "1499392477178439" but the folder with results contains ~1100 images.
Similar is true for val.txt(1 listed, ~50 results).
Test.txt however contains all 120 measurements and results.

Is this an unintentional error in the files on the google drive and is it possible to update it with the full list of files used in training/validation?

All the best

On Windows 10, Anaconda 3 I am currently experiencing the following issue:

Downloading 1EJqr4pvoCE6VahNcoQB9HFccjMMzzc9a into ./kinect/list/motion_real.txt... Done.
Unzipping...C:\Users\<myusername>\AppData\Roaming\Python\Python36\site-packages\google_drive_downloader\google_drive_downloader.py:78: UserWarning: Ignoring `unzip` since "1EJqr4pvoCE6VahNcoQB9HFccjMMzzc9a" does not look like a valid zip file
  warnings.warn('Ignoring `unzip` since "{}" does not look like a valid zip file'.format(file_id))
Downloading 1XXhu7F8Whz_vxOEk4HUAzlIeMQ8E-XzM into ./kinect/list/motion_real/1519695965359234.png... Done.
Unzipping...C:\Users\<myusername>\AppData\Roaming\Python\Python36\site-packages\google_drive_downloader\google_drive_downloader.py:78: UserWarning: Ignoring `unzip` since "1XXhu7F8Whz_vxOEk4HUAzlIeMQ8E-XzM" does not look like a valid zip file
  warnings.warn('Ignoring `unzip` since "{}" does not look like a valid zip file'.format(file_id))

The list goes on for all attempted downloads & unzips. This happens also when running with administrator rights. I've tried debugging in Pycharm and the result is as follows.

Traceback (most recent call last):
  File "C:\Program Files\JetBrains\PyCharm Community Edition 2019.1\helpers\pydev\_pydevd_bundle\pydevd_exec2.py", line 3, in Exec
    exec(exp, global_vars, local_vars)
  File "<input>", line 1, in <module>
  File "C:\Users\<myusername>\.conda\envs\tf190\lib\zipfile.py", line 1131, in __init__
    self._RealGetContents()
  File "C:\Users\<myusername>\.conda\envs\tf190\lib\zipfile.py", line 1198, in _RealGetContents
    raise BadZipFile("File is not a zip file")
zipfile.BadZipFile: File is not a zip file

I've checked that the files cannot be opened even when assigning the .zip extension, with external software. Is there any reason why the script wouldn't run on a Windows platform?

Hi, because I want to generate the data using my camera settings, so i need to download the trans_render, but when i doload the trans_render file using 'python init.py -c trans_render', i found there are too much file whose size only 2Kb, there must be some questions, but i didnot solve it, can you help me to solve it?

Hi, i am very interesting about your project, but you have not gave the trans_render.exe to generate the .pickle file of one scene?

As shown in an example you gave in FLAT\kinect\list\test. According to my understanding, the images with labels that full, ideal, noise, reflection are depth map. The overall color difference between image with label noise and image with label reflection is due to the different scales. An image with label Base err is the difference between labeled depth map and depth groundtruth.
But by reading your released source code, I realize that ideal is the true raw measurement with 9 channels. So I want to ask the meaning of the four labels that full, ideal, noise, reflection. Thanks very much!

Thank you for this dataset! When running python init.py -c trans_render, the first ~20 pickle files download fine, but then all subsequent files downloaded are just Google Drive's html page saying "We're sorry... but your computer or network may be sending automated queries. To protect our users, we can't process your requests right now." It seems that Google Drive blocks mass downloading.

Is there a workaround for this? Perhaps a Google Drive link to the entire trans_render folder that we can download or maybe the multiple pickle files can be zipped into a few large zip files? That would be greatly appreciated, thank you!

Hello,

I downloaded the data from the google drive link given in another issue #11 , but now I had some problems loading the measurements and understanding which of the nine measurements is for which frequency at which phase offset.

First, I loaded the raw kinect correlations as described using the given code:

with open(filename,'rb') as f:
	meas=np.fromfile(f, dtype=np.int32)
meas = np.reshape(meas,(424,512,9)).astype(np.float32)

However the correlations are not correct and looked like this: (small snippet of one image)

[[-54. -51. -50. -68. -55. -59. -57. -46. -47. -50.]
 [ 45.  39.  45.  60.  45.  55.  49.  52.  54.  60.]
 [-47. -51. -51. -49. -52. -54. -48. -47. -57. -66.]
 [ 49.  48.  49.  62.  48.  46.  41.  41.  48.  39.]
 [-50. -53. -55. -52. -36. -54. -50. -53. -40. -48.]
 [ 56.  45.  52.  53.  50.  65.  54.  49.  41.  47.]
 [-37. -50. -48. -50. -57. -50. -50. -51. -53. -42.]
 [ 56.  50.  56.  48.  45.  49.  46.  52.  43.  44.]
 [-50. -50. -54. -57. -57. -54. -42. -37. -50. -52.]
 [ 67.  54.  42.  43.  56.  45.  51.  45.  50.  43.]
 [-54. -40. -53. -42. -41. -41. -47. -48. -53. -49.]
 [ 50.  48.  43.  44.  46.  51.  57.  45.  32.  34.]
 [-47. -60. -41. -37. -48. -42. -48. -43. -45. -45.]
 [ 47.  49.  44.  40.  43.  46.  44.  48.  50.  44.]
 [-47. -37. -49. -52. -51. -43. -40. -46. -41. -49.]
 [ 40.  44.  57.  48.  35.  62.  34.  51.  57.  49.]
 [-52. -44. -45. -50. -44. -53. -44. -44. -40. -51.]
 [ 38.  51.  45.  40.  42.  46.  47.  44.  39.  46.]
 [-51. -51. -57. -47. -45. -49. -41. -45. -46. -42.]
 [ 39.  45.  42.  42.  51.  42.  48.  46.  37.  38.]]

Every other line seems to have the wrong sign.

To get something that looks like the image in the paper I used this transformation:

sign_fix = np.concatenate(
    (np.tile([1, -1], shape[0] // 4), np.tile([-1, 1], shape[0] // 4)), axis=0
                            ).reshape([ -1, 1, 1])

  meas= meas* sign_fix

Could you verify whether this is the right thing to do, or how to transform the data correctly?

Also could you provide the frequencies and phase_offsets for the measurements?
I was not able to find them in the paper or the supplementary.

So far I think the first 3 are at 40MHz(T=2.5e-8s) and the last three at ~58MHz (T=1.7e-8s) .
(taken from sim/tof_class/cam_real_mult)
But the middle one (low frequency signal) I did not find the frequency in the your code.

The phase offsets I figured should be [240, 120, 0]. Is this correct?

I tried with [0, 120, 240] first, but then the ToF-depth reconstruction is off and decreases with object distance.

I would be happy if you could help matching frequency, / phase_offsets to the data.

Thanks and best regards,

Hi Qi,

We are seeing very low amplitude values near edges in the static scene. We are not sure if it is expected behavior. Would it be possible to give a reason for that? Here are the details of the issue.

Scene number: 1569449449898769
image dimension: 424*512 (Kinect 2 dimension )
Measurement type: ideal (collected the first received signal and ignoring rest of the signal at pixel level )
The low amplitude region is highlighted with red color in the attached screenshot. Typically the value ranges from 10-5 to 10-7

Hi, I am currently having an issue with downloading transient files.

Actually I can run the codes for downloading,
(i.e. python init.py -c trans_render)

however all the tranient files are now are 2KB, and I cant even open it (something maybe corrupted)

is it possible to share the google drive link that I can download them manually?

thanks in advance for your help!

Hi Qi,
Could you please let me know which normalization technique is used for the inputs (raw measurements ) while training the neural networks (MoM/ MRM/ MoM and MRM).

Through run python init.py -n all, I can only download the raw measurement of Kinect. So I ran python init.py -c phasor and python init.py -c deeptof next to verify that I could download Phasor and DeepToF, but nothing happened. So I want to make sure that you have provided the raw measurement of Phasor and DeepToF data that can download.