:closed_book: Inference

This document contains instructions for running the READMem variants (i.e., MiVOS, STCN, QDMN) on the DAVIS (D17) and Long-Video (LV1) datasets.

Prerequisites:

• A set workspace as described in Installation.md
• :hotsprings: The D17 and LV1 datasets downloaded.

---

🟥 READMem-MiVOS

• Most useful arguments IMO. But use the python READMem_MiVOS.py -h for more helpful details.

  python READMem_MiVOS.py \
      --model [path/to/the/MiVOS/saves/propagation_model.pth] \
      --output [output/folder/name] \
      --dataset [dataset_name (either D17 or LV1)] \
      --split [by default on val] \
      --mem_confi [path/to/the/memory/configuration] \
      --silence [to avoid icecream statements]
  

• A working command from the get-go, if everything is set correctly:

  python Test_READMem/READMem_MiVOS.py \
      --output OUTPUT_TEST \
      --dataset D17 \
      --silence
  

---

🟦 READMem-STCN

• General usage: (use the -h argument for more details)

  python READMem_STCN.py \
      --model [path/to/the/STCN/saves/STCN.pth] \
      --output [output/folder/name] \
      --dataset [dataset_name (either D17 or LV1)] \
      --split [by default on val] \
      --mem_confi [path/to/the/memory/configuration] \
      --silence [to avoid icecream statements]
  

• Command from the get-go:

  python Test_READMem/READMem_STCN.py \
      --output OUTPUT_TEST_STCN \
      --dataset D17 \
      --silence
  

---

🟧 READMem-QDMN

• General usage: (use the -h argument for more details)

  python READMem_QDMN.py \
      --model [path/to/the/QDMN/saves/QDMN.pth] \
      --output [output/folder/name] \
      --dataset [dataset_name (either D17 or LV1)] \
      --split [by default on val] \
      --mem_confi [path/to/the/memory/configuration] \
      --silence [to avoid icecream statements]
  

• Command from the get-go:

  python Test_READMem/READMem_QDMN.py \
      --output OUTPUT_TEST_QDMN \
      --dataset D17 \
      --silence