Spaces:
Sleeping
Sleeping
Update README.md
#1
by VectorV17 - opened
README.md
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
---
|
| 2 |
-
title:
|
| 3 |
-
emoji:
|
| 4 |
colorFrom: pink
|
| 5 |
colorTo: blue
|
| 6 |
sdk: gradio
|
|
@@ -8,44 +8,3 @@ sdk_version: 6.11.0
|
|
| 8 |
app_file: app.py
|
| 9 |
pinned: false
|
| 10 |
---
|
| 11 |
-
AI Tech Pack Generator For The Garment Industry
|
| 12 |
-
|
| 13 |
-
Inspiration for the Project
|
| 14 |
-
Conceptual Learning: The project is inspired by the efficiency of Stable Diffusion in learning specific, high-detail concepts, as showcased in the source notebook.
|
| 15 |
-
|
| 16 |
-
Automation of Blueprints: In a garment context, the inspiration comes from the need to replace expensive, manual technical sketching with automated AI that understands geometric "blueprints".
|
| 17 |
-
|
| 18 |
-
Rapid Prototyping: The project seeks to solve the slow "Sketch-to-Manufacturer" cycle by allowing designers to iterate on technical flats using text prompts rather than manual CAD software.
|
| 19 |
-
|
| 20 |
-
Structure and Working
|
| 21 |
-
Data Pipeline: The system follows the notebook's structure of Preprocessing Images (resizing to 512x512 and normalizing pixels to a [-1, 1] range) and Assembling Text Datasets containing descriptive prompts.
|
| 22 |
-
|
| 23 |
-
Textual Inversion: A unique placeholder_token (such as <shirt-tech-pack>) is initialized and added to the model’s vocabulary to represent the specific garment style.
|
| 24 |
-
|
| 25 |
-
Selective Fine-Tuning: Similar to the notebook's "StableDiffusionFineTuner," only the Text Transformer (Text Encoder) is set as trainable, while the diffusion model and decoder remain "locked" to preserve base knowledge.
|
| 26 |
-
|
| 27 |
-
Parameter Conditioning: By zipping technical images with varied text descriptions (e.g., "A tech pack for a shirt with a checkered pattern"), the model learns to associate specific aesthetic and physical parameters with the generated image.
|
| 28 |
-
|
| 29 |
-
Accomplishments of the Project
|
| 30 |
-
Automated Flat Generation: Successfully generated 2D technical flats that maintain structural consistency while varying colors (VIBGYOR) and patterns (checkered, geometric).
|
| 31 |
-
|
| 32 |
-
Parametric Scaling: Achieved visual modulation of garment dimensions, such as "chest width," purely through text-based prompt engineering.
|
| 33 |
-
|
| 34 |
-
Cost Efficiency: Reduced the time for producing a basic tech pack from hours of manual labor to under ten minutes of AI generation.
|
| 35 |
-
|
| 36 |
-
What we learned and what's next scope for the project
|
| 37 |
-
What we learned:
|
| 38 |
-
|
| 39 |
-
Precision Matters: Technical drawings require higher fidelity in "Latent Space" than artistic images to ensure lines are straight and measurements appear proportional.
|
| 40 |
-
|
| 41 |
-
Optimizer Stability: Using an Adam optimizer with a CosineDecay learning rate is critical for preventing the model from "forgetting" general shirt structures during training.
|
| 42 |
-
|
| 43 |
-
Next Scope:
|
| 44 |
-
|
| 45 |
-
3D Integration: Extending the 2D tech pack generation into 3D cloth simulations for virtual fitting.
|
| 46 |
-
|
| 47 |
-
Automated BOM: Developing a multi-modal agent that generates the Bill of Materials (text-based list of fabrics/buttons) simultaneously with the technical image.
|
| 48 |
-
|
| 49 |
-
Real-time Collaboration: Building an interface for designers to "tweak" specific seams or stitches in real-time using ControlNet guidance.
|
| 50 |
-
|
| 51 |
-
Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
|
|
|
|
| 1 |
---
|
| 2 |
+
title: E-commerce Intent Evaluation Dashboard
|
| 3 |
+
emoji: ✅️
|
| 4 |
colorFrom: pink
|
| 5 |
colorTo: blue
|
| 6 |
sdk: gradio
|
|
|
|
| 8 |
app_file: app.py
|
| 9 |
pinned: false
|
| 10 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|