inference code not working

[karan-shr]

Hi,
I wanted to test the code by running inference on a couple of samples images i downloaded from the internet. But it's not working. Here are the steps that i followed:

1. cloned repo and setup the conda environment. Note: since there were compatibility issues between tf 1.13 and python 3.5, I used python version 3.6.
2. created a subdirectory in the repo called data/pretrained_weights and downloaded the resnet50 related ckpt files from GDrive into the directory. That is, the files resnet50_v1d.ckpt.data-00000-of-00001 , resnet50_v1d.ckpt.index and resnet50_v1d.ckpt.meta.
3. ran the inference code on a folder containing the following two images:
   1. image_1.
   2. image_2.
4. the images generated in the destination folder don't contain any bounding boxes.

Am I doing something wrong? Could you please guide. Thanks in advance.

[yangxue0827]

pretrained_weights is used for training. you need download trained weights in MODEL_ZOO.md @karan-shr

[karan-shr]

@yangxue0827 thanks for the reply.

Okay so this is what I did:

1. Downloaded the weights for for R3Det with ResNet50_v1d backbone from the link provided in MODELZOO.md doc. Extracted the files and placed them under the output/trained_weights folder. The four files are

   1. checkpoint
   2. DOTA_1079996model.ckpt.data-00000-of-00001
   3. DOTA_1079996model.ckpt.index
   4. DOTA_1079996model.ckpt.meta

2. The data/pretrained_weights contain all the files provided in your Gdrive link mentioned in point 2 of your documentation.

3. The tools/inference.py script uses the standard config file (configs/cfgs.py), so I didn't modify that. Then I ran the inference.py script again and then I got the same result as last time, no bounding boxes.

[yangxue0827]

refer here
@karan-shr

[yangxue0827]

@karan-shr

[yangxue0827]

I updated inference.py and you need to download the script again. @karan-shr

[karan-shr]

Hi @yangxue0827,

Based on your feedback, did the following:

1. Did a git-pull to get the latest version of the inference.py and other changes into the my local repo.
2. changed the folder structure for the weights as you have mentioned in the figure above.

Sadly, still no bounding boxes. Can you please test on the images i had mentioned in the first comment. Just to compare.

Thanks for the feedback so far.

Regards,
Karan

[yangxue0827]

my model is trained by DOTA dataset. which is very different from your images, so no results are normal. Maybe you need to retrain a model with your own data. @karan-shr

[karan-shr]

Yes, i know the images are different. But in this case the bounding box should be wrongly placed, or over- or under-segment. But getting no bounding boxes on an image is also not normal.

Ofcourse, I can retrain the model. But first to determine the efficacy of the code, I am trying your inference code with some simple examples. It should work, shouldn't it? Does the inference script work for you?

[yangxue0827]

Why not use the dota dataset for inference and use the test_dota_r3det.py script.

[ncdejito]

applied tools/inference.py on DOTA Test set 2 images, still no bounding boxes. Adjusted libs/configs/cfgs.py VIS_SCORE to 0.0, still no bounding boxes.

Here's the code I used
python inference.py --data_dir '/dir/of/2/test/images/' --save_dir '/my/chosen/folder/' --gpu 0 --draw_imgs

Here's the output running inference script on DOTA Test set P0034.png

[2018loveyyq]

I also encountered the same problem when running test_dota_r3det.py, and I have placed the file location according to the author's instructions, has your problem been solved?

[2018loveyyq]

I printed out the test results and found that they were all empty