downsample driver produces repeating patterns about tensorstore HOT 8 CLOSED

Can you identify whether it is present in 0.1.52 and/or 0.1.53?

from tensorstore.

Both have the issue. 0.1.51 is the last version that works on my machine.

from tensorstore.

Thanks. I suspect that there's a boundary condition in c6807c4 which triggers this.

from tensorstore.

What is your downsampling for this?

from tensorstore.

Do you mean the downsample_factors? It was [4, 4, 1, 1], and 'mean' was the downsample_method.

from tensorstore.

We are having trouble reproducing this --- can you provide the exact steps to reproduce the issue?

from tensorstore.

here is a minimal reproducible example:

import tensorstore as ts
import cv2

src_spec = {
    'driver':'neuroglancer_precomputed',
    'kvstore':'gs://yuelong-test-data/tensorstore_test/W05_S001R2'
    }
ds_spec = {
    "driver": "downsample",
    "downsample_factors": [4,4,1,1],
    "downsample_method": 'mean',
    "base": src_spec
    }
ds_ts = ts.open(ds_spec).result()
img = ds_ts[4096:4608,4096:4608,0,0].read().result()
cv2.imwrite('test.png', img)

test.png:

The source bucket was created by Michal under lcht-goog-connectomics project, so you should have access to it. Tested on Rocky Linux 8.9 with Python 3.12.2 & Tensorstore 0.1.54.

from tensorstore.

I can verify a repro using examples/extract_slice and a downsampling of [2,2,1,1] or [4,4,1,1], with an output shape of [64, 64, 1, 1]:

SPEC='--input_spec={
  "base": {
    "driver": "neuroglancer_precomputed",
    "kvstore": {
      "bucket": "yuelong-test-data",
      "driver": "gcs",
      "path": "tensorstore_test/W05_S001R2/"
    },
    "transform": {
      "input_exclusive_max": [46696, 50835, 1, 1],
      "input_inclusive_min": [0, 0, 0, 0],
      "input_labels": ["x", "y", "z", "channel"]
    }
  },
  "downsample_factors": [2, 2, 1, 1],
  "downsample_method": "mean",
  "driver": "downsample",
  "dtype": "uint8",
  "transform": {
    "input_exclusive_max": [4672, 4672, 1, 1],
    "input_inclusive_min": [4608, 4608, 0, 0],
    "input_labels": ["y", "x", "z", "channel"]
  }
}'


/tmp/extract_slice.before  --output_file=/tmp/before.png  "${SPEC}"
/tmp/extract_slice.after  --output_file=/tmp/after.png  "${SPEC}"

compare -compose src /tmp/before.png /tmp/after.png /tmp/compare.png

from tensorstore.