System Info
transformers version: current main
Affected file: src/transformers/pipelines/mask_generation.py
Who can help?
No response
Information
Tasks
Description
I was looking into the MaskGenerationPipeline and noticed that when you set a points_per_batch value that doesn't divide evenly into the total number of grid points, the pipeline quietly drops the results from the last batch — no error, no warning, just missing masks.
The root cause is this line in preprocess:
is_last = i == n_points - points_per_batch
eg: n_points=100, points_per_batch=64. The loop runs at i=0 and i=64. At i=64, the check asks 64 == 100-64 which is 64 == 36 — always False. So the final batch never gets flagged as the last one.
The pipeline's PipelinePackIterator relies on this is_last flag to know when to stop accumulating results. When it never sees is_last=True, it calls next() on an already-finished generator, hits StopIteration, and exits — leaving the last batch's masks on the floor.
With SAM's default point grid, n_points is rarely a round multiple of the default points_per_batch=64, so this silently affects most real-world usage.
Reproduction
from transformers import pipeline
from PIL import Image
import requests
image = Image.open(requests.get("http://images.cocodataset.org/val2017/000000039769.jpg", stream=True).raw)
generator = pipeline("mask-generation", model="facebook/sam-vit-base")
# points_per_batch=50 causes n_points % points_per_batch != 0 for typical grids
outputs_partial = generator(image, points_per_batch=50)
outputs_full = generator(image, points_per_batch=None) # all at once, no batching
# outputs_partial["masks"] will have fewer masks than outputs_full["masks"]
print(len(outputs_partial["masks"]), "vs", len(outputs_full["masks"]))Expected behavior
All generated masks should be returned regardless of whether n_points is a multiple of points_per_batch.
System Info
transformers version: current main
Affected file: src/transformers/pipelines/mask_generation.py
Who can help?
No response
Information
Tasks
examplesfolder (such as GLUE/SQuAD, ...)Description
I was looking into the MaskGenerationPipeline and noticed that when you set a points_per_batch value that doesn't divide evenly into the total number of grid points, the pipeline quietly drops the results from the last batch — no error, no warning, just missing masks.
The root cause is this line in preprocess:
is_last = i == n_points - points_per_batcheg: n_points=100, points_per_batch=64. The loop runs at i=0 and i=64. At i=64, the check asks 64 == 100-64 which is 64 == 36 — always False. So the final batch never gets flagged as the last one.
The pipeline's PipelinePackIterator relies on this is_last flag to know when to stop accumulating results. When it never sees is_last=True, it calls next() on an already-finished generator, hits StopIteration, and exits — leaving the last batch's masks on the floor.
With SAM's default point grid, n_points is rarely a round multiple of the default points_per_batch=64, so this silently affects most real-world usage.
Reproduction
Expected behavior
All generated masks should be returned regardless of whether n_points is a multiple of points_per_batch.