unstructuredio/SCORE-Bench · Datasets at Hugging Face

text stringlengths 0 2.18k
--------------------------------------------------- Unstructured Table Begin
Model Latent size Num Z Train steps FID ↓ SSIM ↑ PSNR ↑ L1 ↓
ViT-VQGAN* 32x32 8192 500,000 1.28 - - -
RQ-VAE* 8x8x16 16384 10 epochs 1.83 - - -
Mo-VQGAN* 16x16x4 1024 40 epochs 1.12 0.673 22.42 -
VQ CompVis 32x32 16384 971,043 1.34 0.650 23.85 0.0533
KL CompVis 32x32 - 246,803 0.968 0.692 25.11 0.0474
Sber-VQGAN 32x32 8192 1 epoch 1.44 0.682 24.31 0.0503
Sber-MoVQGAN 67M 32x32 1024 5,000,000 1.34 0.704 25.68 0.0451
Sber-MoVQGAN 67M 32x32 16384 2,000,000 0.965 0.725 26.45 0.0415
Sber-MoVQGAN 102M 32x32 16384 2,360,000 0.776 0.737 26.89 0.0398
Sber-MoVQGAN 270M 32x32 16384 1,330,000 0.686 0.741 27.04 0.0393
--------------------------------------------------- Unstructured Table End


the quality of images (FID 9.86 vs 9.87). The best CLIP score and human-eval score are obtained by diffusion prior.


The best FID score is achieved using Linear Prior. This configuration stands out with the best FID score of 8.03. It is an intriguing outcome: the simplest linear mapping showcased the best FID, suggesting that there might exist a linear relationship between visual and textual embedding vector spaces. To further scrutinize this hypothesis, we trained a linear mapping on a subset of 500 cat images and termed it the "cat prior". Astonishingly, this mapping displayed high proficiency (cf. Figure 5).


--------------------------------------------------- Unstructured Sub-Title Begin
7 Conclusion
--------------------------------------------------- Unstructured Sub-Title End


We presented Kandinsky, a system for various image generation and processing tasks based on a novel latent diffusion model. Our model yielded the SotA results among open-sourced systems. Additionally, we provided an extensive ablation study of an image prior to design choices. Our system is equipped with free-to-use interfaces in the form of Web application and Telegram messenger bot. The pre-trained models are available on Hugging Face, and the source code is released under a permissive
license enabling various, including commercial, applications of the developed technology.


In future research, our goal is to investigate the potential of the latest image encoders. We plan to explore the development of more efficient UNet architectures for text-to-image tasks and focus on improving the understanding of textual prompts. Additionally, we aim to experiment with generating images at higher resolutions and to investigate new features extending the model: local image editing by a text prompt, attention reweighting, physics-based generation control, etc. The robustness against generating abusive content remains a crucial concern, warranting the exploration of real-time moderation layers or robust classifiers to mitigate undesirable, e.g. toxic or abusive, outputs.


--------------------------------------------------- Unstructured Sub-Title Begin
8 Limitations
--------------------------------------------------- Unstructured Sub-Title End


The current system produces images that appear natural, however, additional research can be conducted to (1) enhance the semantic coherence between the input text and the generated image, and (2) to improve the absolute values of FID and image quality based on human evaluations.
--------------------------------------------------- Unstructured Plain Text Format 1.0.4
--------------------------------------------------- Unstructured Title Begin
Answers To Selected Teaching Activities
--------------------------------------------------- Unstructured Title End


--------------------------------------------------- Unstructured Sub-Title Begin
Activity 1: Renewable Energy Basics
--------------------------------------------------- Unstructured Sub-Title End
1. Definitions, examples, and specific advantages and disadvantages are listed in the Facts About Renewable Energy section.
2. The primary fossil fuel energy sources are petroleum, natural gas, and coal.
3. The main advantage of fossil fuels is that they are relatively abundant, and therefore, relatively inexpensive.
4. The primary disadvantage of fossil fuels is that they are more polluting than renewable energy sources. The burning of fossil fuels also produces carbon dioxide, which some fear is causing global warming. This, however, is only a theory, and has not been confirmed by scientific evidence.
--------------------------------------------------- Unstructured Sub-Title Begin
Activity 2: Graphing Energy Facts
--------------------------------------------------- Unstructured Sub-Title End




--------------------------------------------------- Unstructured Sub-Title Begin
Part A
--------------------------------------------------- Unstructured Sub-Title End
1. Make sure students have neatly labeled, colored graphs.
2. Primary energy sources are basic sources of energy, such as coal, natural gas, hydropower, wind, petroleum, etc. Secondary sources, such as electricity, require primary sources of energy to generate power.
3. Petroleum (46.9 percent to 40.4 percent). This large decrease occurred because the price of oil increased significantly in the 1970s. As price increased, consumers bought less, switched to substitutes, etc.
4. Nuclear power. Nuclear power is clean and relatively cost effective. While much of the increased capacity in nuclear power prior to the 1970s was already planned, the oil price increases certainly encouraged the increased use of nuclear power. However, the Three Mile Island incident in 1979 caused much public opposition to nuclear energy. Since then no new plants have been ordered. The growth in the amount of nuclear generated electrical power has tapered off in recent years and could possibly diminish in the future, as older power plants are retired. The future looks brighter for coal, although the current fear of global warming is causing second thoughts about relying more and more on coal.
5. The major reason is that, compared to other sources of energy, renewable sources are relatively more expensive.
--------------------------------------------------- Unstructured Sub-Title Begin
Part B:
--------------------------------------------------- Unstructured Sub-Title End
1. Make sure student graphs are neatly labeled and use several colors.
2. Hydropower: 85.9 percent, Geothermal 3.0 percent, Biomass (Municipal Waste to Energy) 2.3 percent, Biomass (Other, especially wood and wood waste) 6.7 percent, Solar Thermal 0.5 percent, Wind 1.6 percent.
3. Hydropower. It is relatively cost effective compared to the other sources.
4. Solar Thermal
5. Answers will vary.
--------------------------------------------------- Unstructured Sub-Title Begin
Activity 3: Trends In R & D Spending
--------------------------------------------------- Unstructured Sub-Title End
1. Make sure students label the axes correctly. You may have to help students determine the range of R & D on the vertical axis. A workable range is $0 to $900.
--------------------------------------------------- Unstructured Plain Text Format 1.0.4
2. R & D increases rapidly until 1980, then decreases rapidly throughout the 1980s. In 1991, it increases again.
3. Make sure students label the graph correctly and put a workable range of prices on the vertical axis ($0 to $40). You can have students graph the real price changes in oil, too, using 1987 dollars. See teacher directions for this activity.
4. Oil prices rose sharply in the 1970s. They then plunged in the mid-1980s, before increasing again at the end of the decade. The increase in R & D is explained by the observed dramatic rise in oil prices in the 1970s; the decrease in R & D parallels the fall in oil prices. Increases in R & D spending in FY91 can be partly explained by environmental concerns of burning fossil fuels.
--------------------------------------------------- Unstructured Sub-Title Begin
Activity 4: Energy Efficiency
--------------------------------------------------- Unstructured Sub-Title End
1. Energy efficiency measures the amount of energy it takes to do a certain amount of work or do a certain task.
2. Answers will vary. Examples: add insulation, install energy efficient appliances, turn down the thermostat, run dishwashers and washing machines only when fully loaded.
3. Answers will vary. Examples: Improved energy management such as better maintenance, improved insulation, conservation goals, lower thermostats, routine energy audits, use of computers to monitor energy consumption, heat recovery and heat exchange, improvements in electricity cogeneration, investment in energy efficient production technologies.
4. a 6.8 (81.1 - 74.3)
b. 6.8/74.3 = 9.15 percent
c. Energy efficiency has increased greatly.
5. “Energy efficiency” is a commonly used statistic to make comparisons among countries; however, it can be misleading since it does not take into account differences in life styles, population density, industry mix, and other factors. For example, Japan and Italy are small countries with high population densities. This makes energy-saving mass transit more practical. Italy and Japan also tax energy much more heavily (In Italy gas costs about $4 a gallon, of which $3 is tax!), which reduces energy consumption. The United States has a more extreme climate, which requires large amounts of energy for heating and cooling. Living standards also are higher in the United States, and it takes more energy to heat our larger homes. When corrected for differences in living space, the United States is among the most efficient of the other developed countries in residential heating. Another factor is that because energy is relatively abundant in the United States compared to Japan and Italy, we have developed industries that rely on high energy usage (“energy intensity”) in production.
--------------------------------------------------- Unstructured Sub-Title Begin
Activity 8: Case Study: The Case of the Energy Subsidy
--------------------------------------------------- Unstructured Sub-Title End
The Decision Worksheets for the various special interest groups will reflect the biases of the constituencies represented. Nevertheless, the consensus Decision Grid is likely to look something like the sample below.
--------------------------------------------------- Unstructured Plain Text Format 1.0.4
--------------------------------------------------- Unstructured Image Begin

--------------------------------------------------- Unstructured Table Begin

Model Latent size Num Z Train steps FID ↓ SSIM ↑ PSNR ↑ L1 ↓

ViT-VQGAN* 32x32 8192 500,000 1.28 - - -

RQ-VAE* 8x8x16 16384 10 epochs 1.83 - - -

Mo-VQGAN* 16x16x4 1024 40 epochs 1.12 0.673 22.42 -

VQ CompVis 32x32 16384 971,043 1.34 0.650 23.85 0.0533

KL CompVis 32x32 - 246,803 0.968 0.692 25.11 0.0474

Sber-VQGAN 32x32 8192 1 epoch 1.44 0.682 24.31 0.0503

Sber-MoVQGAN 67M 32x32 1024 5,000,000 1.34 0.704 25.68 0.0451

Sber-MoVQGAN 67M 32x32 16384 2,000,000 0.965 0.725 26.45 0.0415

Sber-MoVQGAN 102M 32x32 16384 2,360,000 0.776 0.737 26.89 0.0398

Sber-MoVQGAN 270M 32x32 16384 1,330,000 0.686 0.741 27.04 0.0393

--------------------------------------------------- Unstructured Table End

the quality of images (FID 9.86 vs 9.87). The best CLIP score and human-eval score are obtained by diffusion prior.

The best FID score is achieved using Linear Prior. This configuration stands out with the best FID score of 8.03. It is an intriguing outcome: the simplest linear mapping showcased the best FID, suggesting that there might exist a linear relationship between visual and textual embedding vector spaces. To further scrutinize this hypothesis, we trained a linear mapping on a subset of 500 cat images and termed it the "cat prior". Astonishingly, this mapping displayed high proficiency (cf. Figure 5).

--------------------------------------------------- Unstructured Sub-Title Begin

7 Conclusion

--------------------------------------------------- Unstructured Sub-Title End

We presented Kandinsky, a system for various image generation and processing tasks based on a novel latent diffusion model. Our model yielded the SotA results among open-sourced systems. Additionally, we provided an extensive ablation study of an image prior to design choices. Our system is equipped with free-to-use interfaces in the form of Web application and Telegram messenger bot. The pre-trained models are available on Hugging Face, and the source code is released under a permissive

license enabling various, including commercial, applications of the developed technology.

In future research, our goal is to investigate the potential of the latest image encoders. We plan to explore the development of more efficient UNet architectures for text-to-image tasks and focus on improving the understanding of textual prompts. Additionally, we aim to experiment with generating images at higher resolutions and to investigate new features extending the model: local image editing by a text prompt, attention reweighting, physics-based generation control, etc. The robustness against generating abusive content remains a crucial concern, warranting the exploration of real-time moderation layers or robust classifiers to mitigate undesirable, e.g. toxic or abusive, outputs.

--------------------------------------------------- Unstructured Sub-Title Begin

8 Limitations

--------------------------------------------------- Unstructured Sub-Title End

The current system produces images that appear natural, however, additional research can be conducted to (1) enhance the semantic coherence between the input text and the generated image, and (2) to improve the absolute values of FID and image quality based on human evaluations.

--------------------------------------------------- Unstructured Plain Text Format 1.0.4

--------------------------------------------------- Unstructured Title Begin

Answers To Selected Teaching Activities

--------------------------------------------------- Unstructured Title End

--------------------------------------------------- Unstructured Sub-Title Begin

Activity 1: Renewable Energy Basics

--------------------------------------------------- Unstructured Sub-Title End

1. Definitions, examples, and specific advantages and disadvantages are listed in the Facts About Renewable Energy section.

2. The primary fossil fuel energy sources are petroleum, natural gas, and coal.

3. The main advantage of fossil fuels is that they are relatively abundant, and therefore, relatively inexpensive.

4. The primary disadvantage of fossil fuels is that they are more polluting than renewable energy sources. The burning of fossil fuels also produces carbon dioxide, which some fear is causing global warming. This, however, is only a theory, and has not been confirmed by scientific evidence.

--------------------------------------------------- Unstructured Sub-Title Begin

Activity 2: Graphing Energy Facts

--------------------------------------------------- Unstructured Sub-Title End

--------------------------------------------------- Unstructured Sub-Title Begin

Part A

--------------------------------------------------- Unstructured Sub-Title End

1. Make sure students have neatly labeled, colored graphs.

2. Primary energy sources are basic sources of energy, such as coal, natural gas, hydropower, wind, petroleum, etc. Secondary sources, such as electricity, require primary sources of energy to generate power.

3. Petroleum (46.9 percent to 40.4 percent). This large decrease occurred because the price of oil increased significantly in the 1970s. As price increased, consumers bought less, switched to substitutes, etc.

4. Nuclear power. Nuclear power is clean and relatively cost effective. While much of the increased capacity in nuclear power prior to the 1970s was already planned, the oil price increases certainly encouraged the increased use of nuclear power. However, the Three Mile Island incident in 1979 caused much public opposition to nuclear energy. Since then no new plants have been ordered. The growth in the amount of nuclear generated electrical power has tapered off in recent years and could possibly diminish in the future, as older power plants are retired. The future looks brighter for coal, although the current fear of global warming is causing second thoughts about relying more and more on coal.

5. The major reason is that, compared to other sources of energy, renewable sources are relatively more expensive.

--------------------------------------------------- Unstructured Sub-Title Begin

Part B:

--------------------------------------------------- Unstructured Sub-Title End

1. Make sure student graphs are neatly labeled and use several colors.

2. Hydropower: 85.9 percent, Geothermal 3.0 percent, Biomass (Municipal Waste to Energy) 2.3 percent, Biomass (Other, especially wood and wood waste) 6.7 percent, Solar Thermal 0.5 percent, Wind 1.6 percent.

3. Hydropower. It is relatively cost effective compared to the other sources.

4. Solar Thermal

5. Answers will vary.

--------------------------------------------------- Unstructured Sub-Title Begin

Activity 3: Trends In R & D Spending

--------------------------------------------------- Unstructured Sub-Title End

1. Make sure students label the axes correctly. You may have to help students determine the range of R & D on the vertical axis. A workable range is $0 to $900.

--------------------------------------------------- Unstructured Plain Text Format 1.0.4

2. R & D increases rapidly until 1980, then decreases rapidly throughout the 1980s. In 1991, it increases again.

3. Make sure students label the graph correctly and put a workable range of prices on the vertical axis ($0 to $40). You can have students graph the real price changes in oil, too, using 1987 dollars. See teacher directions for this activity.

4. Oil prices rose sharply in the 1970s. They then plunged in the mid-1980s, before increasing again at the end of the decade. The increase in R & D is explained by the observed dramatic rise in oil prices in the 1970s; the decrease in R & D parallels the fall in oil prices. Increases in R & D spending in FY91 can be partly explained by environmental concerns of burning fossil fuels.

--------------------------------------------------- Unstructured Sub-Title Begin

Activity 4: Energy Efficiency

--------------------------------------------------- Unstructured Sub-Title End

1. Energy efficiency measures the amount of energy it takes to do a certain amount of work or do a certain task.

2. Answers will vary. Examples: add insulation, install energy efficient appliances, turn down the thermostat, run dishwashers and washing machines only when fully loaded.

3. Answers will vary. Examples: Improved energy management such as better maintenance, improved insulation, conservation goals, lower thermostats, routine energy audits, use of computers to monitor energy consumption, heat recovery and heat exchange, improvements in electricity cogeneration, investment in energy efficient production technologies.

4. a 6.8 (81.1 - 74.3)

b. 6.8/74.3 = 9.15 percent

c. Energy efficiency has increased greatly.

5. “Energy efficiency” is a commonly used statistic to make comparisons among countries; however, it can be misleading since it does not take into account differences in life styles, population density, industry mix, and other factors. For example, Japan and Italy are small countries with high population densities. This makes energy-saving mass transit more practical. Italy and Japan also tax energy much more heavily (In Italy gas costs about $4 a gallon, of which $3 is tax!), which reduces energy consumption. The United States has a more extreme climate, which requires large amounts of energy for heating and cooling. Living standards also are higher in the United States, and it takes more energy to heat our larger homes. When corrected for differences in living space, the United States is among the most efficient of the other developed countries in residential heating. Another factor is that because energy is relatively abundant in the United States compared to Japan and Italy, we have developed industries that rely on high energy usage (“energy intensity”) in production.

--------------------------------------------------- Unstructured Sub-Title Begin

Activity 8: Case Study: The Case of the Energy Subsidy

--------------------------------------------------- Unstructured Sub-Title End

The Decision Worksheets for the various special interest groups will reflect the biases of the constituencies represented. Nevertheless, the consensus Decision Grid is likely to look something like the sample below.

--------------------------------------------------- Unstructured Plain Text Format 1.0.4

--------------------------------------------------- Unstructured Image Begin