Fix batch processing for all Human Parser nodes

[wzyfromhust]

Problem

All three Human Parser nodes (ATR, LIP, Pascal) had a critical batch processing bug:

• When using "Load Image Batch From Dir" to load multiple images, only the first image was processed
• All images in the batch received the same parsing result (from image[0])
• This made batch processing completely unusable

Root Cause

The nodes used generate(image[0], ...) which forced processing of only the first image in the batch, ignoring all subsequent images.

Solution

Modified all three nodes to properly iterate over the batch:

• Process each image individually: for index in range(len(image))
• Collect results in lists: ret_masks.append(...) and ret_maps.append(...)
• Concatenate outputs: torch.cat(ret_masks, dim=0)

Files Changed

• HumanParserATRCustomNode.py
• HumanParserLIPCustomNode.py
• HumanParserPascalCustomNode.py
• test_batch_fix.py (added comprehensive test suite)

Testing

✅ All tests passed:

• Single image mode: backward compatible
• Batch mode: each image processed correctly with unique results
• Output shapes: correct tensor dimensions
• Manual testing: verified with real batch workflow

This fix enables proper batch processing workflows in ComfyUI.

[wzyfromhust]

@gabidobo @TCNOco @haohaocreates