90 lines
3.7 KiB
Markdown
90 lines
3.7 KiB
Markdown
# Memory Optimization Plan: Strategy 2 - Load Grayscale Directly
|
|
|
|
This plan outlines the steps to implement memory optimization strategy #2, which involves loading known grayscale map types directly as grayscale images using OpenCV's `IMREAD_GRAYSCALE` flag. This reduces the memory footprint compared to loading them with `IMREAD_UNCHANGED` and then potentially converting later.
|
|
|
|
## 1. Identify Target Grayscale Map Types
|
|
|
|
Define a list of map type names (case-insensitive check recommended during implementation) that should always be treated as single-channel grayscale data.
|
|
|
|
**Initial List:**
|
|
```python
|
|
GRAYSCALE_MAP_TYPES = ['HEIGHT', 'ROUGH', 'METAL', 'AO', 'OPC', 'MASK']
|
|
```
|
|
*(Note: This list might need adjustment based on specific preset configurations or workflow requirements.)*
|
|
|
|
## 2. Modify `_process_maps` Loading Logic
|
|
|
|
Locate the primary image loading section within the `_process_maps` method in `asset_processor.py` (around line 608).
|
|
|
|
**Change:** Before calling `cv2.imread`, determine the correct flag based on the `map_type`:
|
|
|
|
```python
|
|
# (Define GRAYSCALE_MAP_TYPES list earlier in the scope or class)
|
|
|
|
# ... inside the loop ...
|
|
full_source_path = self.temp_dir / source_path_rel
|
|
|
|
# Determine the read flag
|
|
read_flag = cv2.IMREAD_GRAYSCALE if map_type.upper() in GRAYSCALE_MAP_TYPES else cv2.IMREAD_UNCHANGED
|
|
log.debug(f"Loading source {source_path_rel.name} with flag: {'GRAYSCALE' if read_flag == cv2.IMREAD_GRAYSCALE else 'UNCHANGED'}")
|
|
|
|
# Load the image using the determined flag
|
|
img_loaded = cv2.imread(str(full_source_path), read_flag)
|
|
|
|
if img_loaded is None:
|
|
raise AssetProcessingError(f"Failed to load image file: {full_source_path.name} with flag {read_flag}")
|
|
|
|
# ... rest of the processing logic ...
|
|
```
|
|
|
|
## 3. Modify `_merge_maps` Loading Logic
|
|
|
|
Locate the image loading section within the resolution loop in the `_merge_maps` method (around line 881).
|
|
|
|
**Change:** Apply the same conditional logic to determine the `imread` flag when loading input maps for merging:
|
|
|
|
```python
|
|
# ... inside the loop ...
|
|
input_file_path = self.temp_dir / res_details['path']
|
|
|
|
# Determine the read flag (reuse GRAYSCALE_MAP_TYPES list)
|
|
read_flag = cv2.IMREAD_GRAYSCALE if map_type.upper() in GRAYSCALE_MAP_TYPES else cv2.IMREAD_UNCHANGED
|
|
log.debug(f"Loading merge input {input_file_path.name} ({map_type}) with flag: {'GRAYSCALE' if read_flag == cv2.IMREAD_GRAYSCALE else 'UNCHANGED'}")
|
|
|
|
# Load the image using the determined flag
|
|
img = cv2.imread(str(input_file_path), read_flag)
|
|
|
|
if img is None:
|
|
raise AssetProcessingError(f"Failed to load merge input {input_file_path.name} with flag {read_flag}")
|
|
|
|
# ... rest of the merging logic ...
|
|
```
|
|
|
|
## 4. Verification
|
|
|
|
During implementation in `code` mode:
|
|
* Ensure the `GRAYSCALE_MAP_TYPES` list is defined appropriately (e.g., as a class constant or module-level constant).
|
|
* Confirm that downstream code (e.g., stats calculation, channel extraction, data type conversions) correctly handles numpy arrays that might be 2D (grayscale) instead of 3D (BGR/BGRA). The existing code appears to handle this, but it's important to verify.
|
|
|
|
## Mermaid Diagram of Change
|
|
|
|
```mermaid
|
|
graph TD
|
|
subgraph _process_maps
|
|
A[Loop through map_info] --> B{Is map_type Grayscale?};
|
|
B -- Yes --> C[imread(path, GRAYSCALE)];
|
|
B -- No --> D[imread(path, UNCHANGED)];
|
|
C --> E[Process Image];
|
|
D --> E;
|
|
end
|
|
subgraph _merge_maps
|
|
F[Loop through resolutions] --> G[Loop through required_input_types];
|
|
G --> H{Is map_type Grayscale?};
|
|
H -- Yes --> I[imread(path, GRAYSCALE)];
|
|
H -- No --> J[imread(path, UNCHANGED)];
|
|
I --> K[Use Image in Merge];
|
|
J --> K;
|
|
end
|
|
|
|
style B fill:#f9f,stroke:#333,stroke-width:2px
|
|
style H fill:#f9f,stroke:#333,stroke-width:2px |