david0tt/deepgtav

A system to easily extract ground truth training data for different machine learning tasks from GTAV

deepgtav's Issues

Installation procedure

How can I install the GTAV?
Thank you very much

understanding camera rotation

Hi,

I am trying to understand how to interpret the camera orientation parameters: camrot_x, camrot_y, camrot_z.
Is the camera fixed to the drone (i.e. there is no gimbal)?
Which axis is the view direction?

For example, in this case the z axis is view direction:

I am asking this because I would like to understand from the metadata when the camera is pointing down or at the horizon.

Thanks!

How can we get GTs or annotations of published dataset?

In the dataset download link you posted, I can only find compressed jpeg images but not corresponding annotation files.
I would like to ask how we should obtain the annotation information of the released dataset?
Hope to get your reply as soon as possible

Step 3 install

HI

can you explain the exact location of gtav install directory?

you told us to copy the contents of DeepGTAV-preSIL/bin/Release/ to GTAV install directory

is this here?

I want to use it to generate infrared data.

Would appreciate your help on whether it is possible to use in-game IR items, etc.? I'm guessing that it seems to be possible but have no clue

I am curious about the format of the metadata.

Hello.
Thank you for the good research results you showed using synthetic data.

I would like to ask for information about the format of metadata.

The pre-generated dataset you posted was very helpful to me.
Thank you.

<Sample of meta data>
a: -425.1471252441406
b: -559.9451293945312 
c: 77.26197814941406 
d: 50.85673904418945 
e: -427.5518493652344 
f: -558.969970703125 
g: 57.43088912963867 
h: -81.8794174194336 
i: -6.996013164520264 
j: 137.6229705810547 
k: 17 
l: 0 
m: 36 
n: EXTRASUNNY

Stationary Camera Option

Hey,

I want to generate data from the drone perspective and have the drone stay at the same location. Similar to a stationary security camera.
I've tried modifying datageneration_VisDrone.py (disabled all the GoToLocation messages), but the drone always flies itself forward.

Is there an option to force the drone from moving?

Thanks!

mapping category id to class name seadronesee

What is the mapping between category id and the class name?
I am using the same one as that of Seadronesee (real) but it seems not working.

This is the one taken from Seadronesee:

Thanks

road semgentation and multiple side-view cameras

Hello,

Great work!

I have two questions:

1. Do you think it would be possible to obtain also road segmentation (e.g., drivable/non drivable)?
2. Do you think it would be possible to mount multiple side-view cameras on top of a car while also getting the aerial view?

Thanks!

Hi, we have already get some screenshot by python scripts. However, the Bounding box is still not working. I think the main problem comes from message part [message = client.recvMessage()] and **message['bbox2d'] always equal to None**. Do you have any suggestion about this problem? Thanks for your reply

          Hi, we have already get some screenshot by python scripts. However, the Bounding box is still not working. I think the main problem comes from message part [message = client.recvMessage()] and **message['bbox2d'] always equal to None**. Do you have any suggestion about this problem? Thanks for your reply

Originally posted by @RuiCheng19950208 in #10 (comment)

If the DGTA-VisDrone datasets can be directly download?

Hi! It is exciting to find such a useful work. I think it is truly important for researchers who fight with UAVs images. In fact, I'm now struggling in finding a virtual UAV datasets for my works with the topic of domain adapation in UAVs. Therefore, It is very exciting for me to find your work. It seems that the DGTA-VisDrone is very nice to me.
Then,
If the DGTA-VisDrone datasets can be directly download? I know the dataset can be generate by myself and Readme.md is also clear. But if the datasets can be directly downloaded, it will be more likely that my work can convinced my supervisor.
Thanks for your amazing work!

image missing in train set jpg data dgta-seadronesee

Hi and thanks for sharing this useful project. Great job!

I've downloaded the dgta-seadronesee data (jpg folder + labels + meta data) and unzipped the files.
Please check the size in the respective directories.

Train

Val

The images directory of the train set is one item short.

Can you please check which file is missing?

I am going to create a script to convert the annotations in coco format, I guess I have to resize the original bboxes if I want to use them with the jpg images, right?

Folder got generated but not able to see any images, metadata and lables.

I ran the data generation, it is working fine, folder got created in the path and the count is going on but data is not getting generated

The format of DeepGTAV-Visdrone

Are the classes the same as VisDrone?

And I noticed that after the class there are 4 numbers. Is it the same as the yolov5 format? Thanks!

imageSegmentation and 2dbbox received as "None"

Trying the script "presentation_VisDrone.py" everything runs well except for the last part that should create a window (as shown in the Youtube video) to plot the 2Dbbox and segmentation results. This is due to the "None" returned corresponding to the fields "imageSegmentation" and "2dbbox". What may be the reasons?
I have already set MSAA to "OFF". Thanks in advance.

installation

i tried several times but i couldn't do the datageneration using a VPiolot

after this command line python .\VPilot\datageneration_VisDrone.py --save_dir "C:\DeepGTAV_Exportdir"

the game took any effect i cna't see the 2nd window rxtrated data

what can i do ?

Corrupt DGTA_SeaDroneSee/train/images.tar.gz file!

I tried downloading the DGTA_SeaDroneSee/train/images.tar.gz. The file seems to be corrupt.

incorrect bounding boxes some classes seadronesee

There are bounding boxes of some classes that are smaller than the object.
See example below (ignore class name for now, I will open another issue).

These are some zoomed-in cases:

This is the code I am using to visualise this data:

import matplotlib.pyplot as plt
from matplotlib.patches import Rectangle

fns = glob.glob(os.path.join(source_dir, dir, 'labels', '*txt'))
id = 0
for fn in fns:
    imageid = os.path.basename(fn).split('.')[0]
    anns = np.loadtxt(fn, dtype=float)

    img = Image.open(os.path.join(source_dir, dir, 'images', '%s.jpg' % imageid))

    plt.imshow(img)
    ax = plt.gca()
    for i, bbox in enumerate(anns):
        rect = Rectangle((1920 * (bbox[1] - bbox[3]/2),
                          1080 * (bbox[2] - bbox[4]/2)), 1920 * bbox[3], 1080 * bbox[4],
                         linewidth=3,
                         edgecolor='r',
                         facecolor='none')
        plt.text(1920 * (bbox[1] - bbox[3]/2) - 2, 1080 * (bbox[2] - bbox[4]/2) - 2, class_names[int(bbox[0])], fontsize=10, color='r')
        ax.add_patch(rect)
    plt.show()

Can you please check this problem, it apparently occurs only with some classes.