visionlearninggroup / visda-2018-public Goto Github PK

when I submit a result file, this error occurs. but my submission contains 53343 lines which is the same as that image_list.txt file in test folder.

Nice work!
But some initial model required in your code is not provided, such as ssd512_vgg16.pth in this line:

but this score is not the same as the mean accuracy in the leaderboard.

In the rules the COCO (target) training data is not mentioned anywhere. Are we to allowed to train our domain adaptation algorithm on it? Without using the labels of course. Or is it to compare the domain adaptation model to a model trained on the target training dataset?

I have started working with the detection track test data and I have discovered that the following images cannot be read due to having 0 bytes:

1cfb9a9ecd9f7e4c4c1b9eefb355c58d04741521.jpg
3839e4552f8dc074debcae3bb23e3d344fadc2f5.jpg
6720e49dafc58027896c079dcf31c65e623c526c.jpg
89ee19843ddcaf1bcdb12b19e10ca9245658f8da.jpg
97f07801bfd70c8f5c13e99207eeeec244b39e2a.jpg
b42a52991fc7960823246409930153dc80d1475e.jpg
d0c12a8136c25fbba435a478e618303bb7e52158.jpg
efb57473c96028f06c3bc55648797e4089fe363f.jpg
f41b0275980fee7ed76bbd6eb7c0c4758d7c5616.jpg

As a work around, I can skip them for now and produce empty predictions for them in the predictions file.

I have started working with the detection track. This morning I submitted a results file. However, after 2 hours my submission status is still submitted and no scores are shown. I wonder if it's normal and how long does it usually take for the server to evaluate the results. Thanks!

I submit a zip file just now, which strictly followed your instructions, but I got the error message above from the coda competition platform.
I then tried several times and got the same error.
Would it be possible for you to check whether the platform has crashed or not? @MInner @xcpeng @nkaushik1

The detection tract test data set does not contain or have an associated image_list.txt. As a consequence, we cannot determine which file order our prediction files should use.

As a work around, I generated one using the ls command, hoping that its sorting order is the correct one.

Hi, I run the examples/ssd/train_visda.py script after figure out the directory(using visda18 dataset), and I came across several errors as follows:

After I run next time(without any change), I got the following error message:

When debugging, I found tensor operation buggy in the following code:

I'm still puzzled about the dataset folder configuration. Hope the README file of the code coming out soon : )

According to CodaLab, the challenge ends on 22th August.
According to the Task CV homepage, the submission deadline is on 15th August.

Will the submission deadline be shifted back as well or is it mandatory to submit the paper on 15th?

The training synthetic images are all plain grey albedo images. The real images are textured. Any clue on how to carryout training to recognize textured images?

Hi,

I've been having difficult submitting to the detection competition. It appears that my text file format is incorrect, as I get the following error:

Traceback (most recent call last):
  File "/tmp/codalab/tmpxfH0zT/run/program/evaluate.py", line 438, in 
    detection_evaluation(truth_file, source_file, adaptation_file, output_file)
  File "/tmp/codalab/tmpxfH0zT/run/program/evaluate.py", line 363, in detection_evaluation
    for line in f.read().splitlines()
  File "/tmp/codalab/tmpxfH0zT/run/program/evaluate.py", line 21, in chunks
    raise RuntimeError("Sequence does not split into {k}, because last elem has {elem.index(uniq)} left")
RuntimeError: Sequence does not split into {k}, because last elem has {elem.index(uniq)} left

Would it be possible for you guys to either:

• provide an example submission file that the server can handle
• provide code to generate a valid submission file
• provide the source code of 'evaluate.py' that runs on the server
• modify the evaluation code so that it reports more helpful error messages

Thanks! :)

Hello,

I find something wrong with the result board shown at the end the following page: https://competitions.codalab.org/competitions/19113#results

It seems that the board shows the lowest score rather than the highest score?

Best,
Yuming

Thank you for releasing the codes!

https://github.com/VisionLearningGroup/visda-2018-public/blob/master/detection/pytorch-ssd-mmd-coral/train_visda_source.py#L77
Like this, all of the scripts assume the existence of coco17-val.txt.
Is it possible to share coco17-val.txt in addition to coco17-val-names.txt?
Or, should I extract it manually?

Can we use test image when training the domain adaptation model?

Hi,

I have been taking a look at the ground truths that you've provided for the detection competition. I have noticed that the GT boxes in the val_ground_truth.pkl do not match those in coco17-val.txt; it seems that they have been scaled. A consistent x,y scale factor is used for each image, but I am unable to determine the pattern/algorithm used to compute the scale factors. Would it be possible to replace val_ground_truth.pkl with a file generated by direct conversion?

I have created a pull request (#11) that adds a conversion script (in case you're interested) and also changes the detection README slight in order to clarify the first problem that I had concerning format :)

As stated in the title, I cannot find any link to the test data of detection track. And I notice that the ReadMe file in detection folder has not been updated for several days. Could you give me some help for accessing to the test data of detection track? Thanks.

at https://competitions.codalab.org/competitions/19113#results , there are 13 classes but with two horse! and knife is missing......

I want to build my dataset for object detection.
which program is suitable to do that. e.g save bounding box positions , randomize objects...

As reported in the .pdf: https://arxiv.org/pdf/1806.09755.pdf :
"We [...] fully open sourced our 3D model collection"
How can I download just the CAD models, rather than the rendered images?

sorry to miss the ddl.
I have just noticed this:

participants are requested to submit a 2-page abstract directly via email to visda- 
[email protected], **within 1 week of the challenge end.**

can we submit abstract now?

Dear Visda organizers,

 I have not noticed that we need to submit the google form https://goo.gl/forms/G6Kl7OuUVg1BvbRu1  before Aug 25th, I am wonder if i still can submit this form?

Sorry for my carelessness, hope you can give me this chance.

Best Regards,
Qing Lian

Could you please provide a sample submission file? I am not sure how are you calculating the classification accuracy and writing those accuracy in files.