SwanLab

Log Media Data ​

SwanLab supports logging media data (images, audio, text, obejct3d, etc.) to visually explore your experimental results and achieve subjective evaluation of your models.

1. Images ​

swanlab.Image supports logging various image types, including numpy, PIL, Tensor, file reading, etc. API Documentation.

1.1 Log Array-Type Images ​

Array-type includes numpy and tensor. Directly pass the Array into swanlab.Image, and it will automatically handle it according to the type:

  • If it is numpy.ndarray: SwanLab will use pillow (PIL) to read it.
  • If it is tensor: SwanLab will use the make_grid function of torchvision for conversion and then use pillow to read it.

Example code:

python
image = swanlab.Image(image_array, caption="Left: Input, Right: Output")
swanlab.log({"examples": image})

1.2 Log PIL-Type Images ​

Directly pass it into swanlab.Image:

python
image = PIL.Image.fromarray(image_array)
swanlab.log({"examples": image})

1.3 Log File Images ​

Provide the file path to swanlab.Image:

python
image = swanlab.Image("myimage.jpg")
swanlab.log({"example": image})

1.4 Log Matplotlib ​

Pass the plt object of matplotlib.pyplot into swanlab.Image:

python
import matplotlib.pyplot as plt

# Data
x = [1, 2, 3, 4, 5]
y = [2, 3, 5, 7, 11]
# Create a line plot
plt.plot(x, y)
# Add title and labels
plt.title("Examples")
plt.xlabel("X-axis")
plt.ylabel("Y-axis")

swanlab.log({"example": swanlab.Image(plt)})

1.5 Log Multiple Images in One Step ​

Logging multiple images in one step means passing a list composed of swanlab.Image type objects in one swanlab.log.

python
# Create an empty list
image_list = []
for i in range(3):
    random_image = np.random.randint(low=0, high=256, size=(100, 100, 3))
    image = swanlab.Image(random_image, caption=f"Random Image {i}")
    # Add swanlab.Image type objects to the list
    image_list.append(image)

swanlab.log({"examples": image_list})

For more details about images, refer to the API Documentation.

2. Audio ​

API Documentation

2.1 Log Array-Type Audio ​

python
audio = swanlab.Audio(np_array, sample_rate=44100, caption="white_noise")
swanlab.log({"white_noise": audio})

2.2 Log Audio Files ​

python
swanlab.log({"white_noise": swanlab.Audio("white_noise.wav")})

2.3 Log Multiple Audio in One Step ​

python
examples = []
for i in range(3):
    white_noise = np.random.randn(100000)
    audio = swanlab.Audio(white_noise, caption=f"audio_{i}")
    # Add swanlab.Audio type objects to the list
    examples.append(audio)

run.log({"examples": examples})

3. Text ​

API Documentation

3.1 Log Strings ​

python
swanlab.log({"text": swanlab.Text("A example text.")})

3.2 Log Multiple Text in One Step ​

python
# Create an empty list
text_list = []
for i in range(3):
    text = swanlab.Text("A example text.", caption=f"{i}")
    text_list.append(text)

swanlab.log({"examples": text_list})

alt text

4. 3D Point Cloud ​

alt

👀Please refer to this document: API-Oject3D

5. Molecules ​

Please refer to this document: API-Molecule

6. Video ​

Please refer to this document: API-Video

Q&A ​

1. What is the role of the caption parameter? ​

Each media type has a caption parameter, which is used for textual description of the media data. For example, for images:

python
apple_image = swanlab.Image(data, caption="Apple")
swanlab.log({"im": apple_image})

2. How to synchronize media data with the epoch number? ​

When logging media data with swanlab.log, specify the step parameter as the epoch number.

python
for epoch in epochs:
    ···
    swanlab.log({"im": sw_image}, step=epoch)