sulc / tfrecord-viewer

TFRecord Viewer for browsing contents of TFRecords with object detection/classification annotations.

TFRecord Viewer

"How about checking your data before going deeper?"

Use TFRecord Viewer to browse contents of TFRecords with object detection/classification/segmentation annotations.

The viewer runs a Flask server to provide a web gallery with annotation overlays. I.e. you can run it on your server machine, but browse on your local machine.

The web gallery displayed with Fotorama.io.

Examples

python3 tfviewer.py datasets/COCO/tfrecord/coco_train.record-00003-of-00100 --labels-to-highlight='car;truck;bus;motorcycle'

python3 tfviewer.py datasets/imagenet/tfrecord/train-00000-of-01024 --overlay classification

python3 tfviewer.py tfrecord_samples/semantic_cityscapes_train-00000-of-00010.tfrecord --overlay=segmentation --segmap-colormap-file=./cityscapes_colormap.txt

tfrecord-viewer's Issues

“IOError: image file truncated”

Error
In some cases when opening the record file the following error turns up.
OSError: image file is truncated (46 bytes not processed)
Solution
In the file ~/tfrecord-viewer/overlays/detection_overlay.py
Add the following statements to the top
import io from PIL import Image, ImageDraw, ImageFont, ImageFile ImageFile.LOAD_TRUNCATED_IMAGES = True
References
https://stackoverflow.com/questions/12984426/python-pil-ioerror-image-file-truncated-with-big-images

Error when used with Flask 2.2

The flask function 'send_file', used on line 173 of tfviewer.py, doesn't use the 'attachment_filename' keyword as of Flask 2.2. It seems to have been replaced with the 'download_name' keyword.

Draw label over detection boxed

First, thanks for making this. It's super useful!

The Problem

For objects that are very small compared to the total image size, the label might block the object.

I'm currently converting this dataset to TF Records: http://bird.nae-lab.org/dataset/

Suggestion

consider drawing the label over the bounding box.

A Solution

I would have sent a PR but my editor changed the formatting of the file.

Where's what I did:

# file: overlays/detection_overlay.py

...

    def draw_bboxes(self, image_bytes, bboxes):
        """Draw bounding boxes onto image.

        Args:
          image_bytes: JPEG image.
          bboxes (list of tuples): [ (label, xmin, xmax, ymin, ymax), (label, xmin, xmax, ymin, ymax) , .. ]

        Returns:
          image_bytes: JPEG image including bounding boxes.
        """
        img = Image.open(io.BytesIO(image_bytes))

        draw = ImageDraw.Draw(img)

        width, height = img.size

        for bbox in bboxes:
            label, xmin, xmax, ymin, ymax = self.bboxes_to_pixels(bbox, width, height)
            draw.rectangle([xmin, ymin, xmax, ymax], outline=self.bbox_color(label))
            # draw label over bounding box
            w, h = self.font.getsize(label)
            draw.rectangle((xmin, ymin - 2, xmin + w + 4, ymin - h - 4), fill="white")

            draw.text(
                (xmin + 2, ymin - h - 4),
                label,
                fill=self.bbox_color(label),
                font=self.font,
            )

        with io.BytesIO() as output:
            if img.mode in ("RGBA", "P"):
                img = img.convert("RGB")
            img.save(output, format="JPEG")
            output_image = output.getvalue()
        return output_image

Requirements.txt

Got some error in tensorflow no python_io, change to io but still get another error, may be we need to specify the packages version

How to enable Debugging mode

By default Debug mode: off may i know how i can enable this and what are the things i can debug?

Only Rectangle Box Shown

Hi,
Thank you for developing this , but in your viewer file it only takes xmin , ymin , xmax and ymax.But I am doing segmentation so I need to draw bounding box with polygoned segmentation points like below images.

Help me sort out this

boxes are misplaced on the image

Thank you so much for this amazing job! This viewer helps me a lot :)
But I found the boxes are misplaced on the image.

All feature names are default, and all arguments are default.

Thank you so much for making this

No way to thank, so I write it here. Thank you for this amazing job done! +rep

Not working on imagenet image

I ran to error with the followinig error:

python3 ./tfviewer.py /datasets/imagenet2012/5.1.0/imagenet2012-train.tfrecord-00004 --overlay classification

imagenet2012-train.tfrecord-00000-of-01024 imagenet2012-train.tfrecord-00003-of-01024 imagenet2012-train.tfrecord-00006-of-01024 imagenet2012-train.tfrecord-00009-of-01024
imagenet2012-train.tfrecord-00001-of-01024 imagenet2012-train.tfrecord-00004-of-01024 imagenet2012-train.tfrecord-00007-of-01024
imagenet2012-train.tfrecord-00002-of-01024 imagenet2012-train.tfrecord-00005-of-01024 imagenet2012-train.tfrecord-00008-of-01024
root@uefi-popos-20:/tfrecord-viewer/src/tfrecord-viewer# python3 ./tfviewer.py /datasets/imagenet2012/5.1.0/imagenet2012-train.tfrecord-00000-of-01024 --overlay classification
Pre-loading up to 200 examples..
2022-04-14 12:59:53.259380: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:936] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-04-14 12:59:53.264222: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:936] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-04-14 12:59:53.264642: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:936] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-04-14 12:59:53.265321: I tensorflow/core/platform/cpu_feature_guard.cc:151] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations: SSE3 SSE4.1 SSE4.2 AVX AVX2 FMA
To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
2022-04-14 12:59:53.265464: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:936] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-04-14 12:59:53.265895: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:936] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-04-14 12:59:53.266267: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:936] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-04-14 12:59:53.719396: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:936] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-04-14 12:59:53.719827: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:936] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-04-14 12:59:53.720189: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:936] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-04-14 12:59:53.720539: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1525] Created device /job:localhost/replica:0/task:0/device:GPU:0 with 6664 MB memory: -> device: 0, name: NVIDIA GeForce RTX 2070 SUPER, pci bus id: 0000:06:00.0, compute capability: 7.5
Filename: /datasets/imagenet2012/5.1.0/imagenet2012-train.tfrecord-00000-of-01024
Traceback (most recent call last):
File "./tfviewer.py", line 192, in
count = preload_images(args.max_images)
File "./tfviewer.py", line 116, in preload_images
image_bytes = example.features.feature[args.image_key].bytes_list.value[0]
IndexError: list index (0) out of range

How to increase line thickness?

Hello,
I have a tfrecord file which contains bounding boxes of planes.
But my image size is quite large (4800 x 2703). So when I visualize them, bounding boxes look thin.
Therefore, my question is that if there is a way to increase the line thickness of bounding boxes.
I guess it is somewhere in detection_overlay.py but I wasn't able to point it out.

Thank you!

Can you please help me with the command to view tf records generated for detection module tensprflow API.

Throws error: IndexError: list index (0) out of range
python tfviewer.py train_0.record --overlay classification

Throws error:IndexError: list index (0) out of range
python tfviewer.py train_0.record --labels-to-highlight='cat;dog' --overlay classification

I don't install tfrecord-viewer

Not Working For Tensorflow 1 tf record

tfviewer is not working for a model that trained with TensorFlow 1 and it shows the black screen in the browser.

not all images are in tf_record

will this show every image in tf record

Wrapper for tfrecord viewer

First I really want to thank you a lot for this project. It saved me tons of time. Just wanted to let you know that I've written a wrapper around this project to make it more user-friendly with selection feature. You can see it here:
Wrapper for tfrecord viewer