README.md

---
license: apache-2.0
datasets:
- Inst-IT/Inst-IT-Dataset
- lmms-lab/LLaVA-NeXT-Data
language:
- en
metrics:
- accuracy
pipeline_tag: video-text-to-text
tags:
- multimodal
- fine-grained
- instance-understanding
model-index:
- name: LLaVA-Next-Inst-It-Qwen2-7B
  results:
  - task:
      type: multimodal
    dataset:
      name: Inst-IT-Bench-I-OE
      type: Open-Ended
    metrics:
    - type: accuracy
      value: 67.9
      name: accuracy
      verified: true
  - task:
      type: multimodal
    dataset:
      name: Inst-IT-Bench-I-MC
      type: Multi-Choice
    metrics:
    - type: accuracy
      value: 75.3
      name: accuracy
      verified: true
  - task:
      type: multimodal
    dataset:
      name: AI2D
      type: ai2d
    metrics:
    - type: accuracy
      value: 78.7
      name: accuracy
      verified: true
  - task:
      type: multimodal
    dataset:
      name: MMMU
      type: mmmu
    metrics:
    - type: accuracy
      value: 42.7
      name: accuracy
      verified: true
  - task:
      type: multimodal
    dataset:
      name: POPE
      type: pope
    metrics:
    - type: accuracy
      value: 87.6
      name: accuracy
      verified: true
  - task:
      type: multimodal
    dataset:
      name: GQA
      type: gqa
    metrics:
    - type: accuracy
      value: 65.5
      name: accuracy
      verified: true
  - task:
      type: multimodal
    dataset:
      name: MM-Vet
      type: mm-vet
    metrics:
    - type: accuracy
      value: 44.7
      name: accuracy
      verified: true
  - task:
      type: multimodal
    dataset:
      name: Inst-IT-Bench-V-OE
      type: Open-Ended
    metrics:
    - type: accuracy
      value: 45.7
      name: accuracy
      verified: true
  - task:
      type: multimodal
    dataset:
      name: Inst-IT-Bench-V-MC
      type: Multi-Choice
    metrics:
    - type: accuracy
      value: 53.3
      name: accuracy
      verified: true
  - task:
      type: multimodal
    dataset:
      name: ActNet-QA
      type: actnet-qa
    metrics:
    - type: accuracy
      value: 55.2
      name: accuracy
      verified: true
  - task:
      type: multimodal
    dataset:
      name: EgoSchema
      type: egoschema
    metrics:
    - type: accuracy
      value: 50.4
      name: accuracy
      verified: true
  - task:
      type: multimodal
    dataset:
      name: NextQA
      type: nextqa
    metrics:
    - type: accuracy
      value: 73.0
      name: accuracy
      verified: true
  - task:
      type: multimodal
    dataset:
      name: VideoMME
      type: videomme
    metrics:
    - type: accuracy
      value: 54.0
      name: accuracy
      verified: true
  - task:
      type: multimodal
    dataset:
      name: TempoCompass
      type: tempocompass
    metrics:
    - type: accuracy
      value: 63.9
      name: accuracy
      verified: true
---

# LLaVA-Next-Inst-It-Qwen2-7B
[**Homepage**](https://inst-it.github.io/) | [**Code**](https://github.com/inst-it/inst-it) | [**Paper**](https://huggingface.co/papers/2412.03565) | [**arXiv**](https://arxiv.org/abs/2412.03565)

LLaVA-Next-Inst-It-Qwen2-7B is a multimodal model that excels at instance-level understanding, 
which is introduced in the paper [Inst-IT: Boosting Multimodal Instance Understanding via Explicit Visual Prompt Instruction Tuning](https://huggingface.co/papers/2412.03565)

* **Architecture**: siglip-so400m-patch14-384 + Qwen2-7B
* **Data**: LLaVA-NeXT-Data / Inst-IT-Dataset
* **Precision**: bfloat16


## Quick Start
**Install**

Our code is based on LLaVA-NeXT, before running, please install the LLaVA-NeXT to prepare the environment:
```shell
pip install git+https://github.com/LLaVA-VL/LLaVA-NeXT.git
```
**Error Handling**
<details>
<summary>Click to unfold</summary>
  
* **Common error case 1:**
``` shell
  Exception: data did not match any variant of untagged enum ModelWrapper at line 757272 column 3
```
This is caused by the version of `transformers`, try to update it:
``` python
pip install -U transformers
```

* **Common error case 2:**
``` shell
RuntimeError: Error(s) in loading state_dict for CLIPVisionModel:
	size mismatch for vision_model.embeddings.position_embedding.weight: copying a param with shape torch.Size([729, 1152]) from checkpoint, the shape in current model is torch.Size([730, 1152]).
	You may consider adding `ignore_mismatched_sizes=True` in the model `from_pretrained` method.
```
This is a logical error encountered when loading the vision tower from the local path. To fix this issue, you can prepare the environment in any of the following ways.

**Option 1: Install from our fork of LLaVA-NeXT:**

``` shell
pip install git+https://github.com/inst-it/LLaVA-NeXT.git
```

**Option 2: Install from LLaVA-NeXT and manually modify its code:**
* step 1: clone source code
``` shell
git clone https://github.com/LLaVA-VL/LLaVA-NeXT.git
```
* step 2: before installing LLaVA-NeXT, you need to modify `line 17` of [llava/model/multimodal_encoder/builder.py](https://github.com/LLaVA-VL/LLaVA-NeXT/blob/main/llava/model/multimodal_encoder/builder.py#L17).
``` python
# Before modification:
if is_absolute_path_exists or vision_tower.startswith("openai") or vision_tower.startswith("laion") or "ShareGPT4V" in vision_tower:

# After modification:
if "clip" in vision_tower or vision_tower.startswith("openai") or vision_tower.startswith("laion") or "ShareGPT4V" in vision_tower:
```
* step 3: install LLaVA-NeXT from source:
``` shell
cd LLaVA-NeXT
pip install --upgrade pip  # Enable PEP 660 support.
pip install -e ".[train]"
```

We recommend the first way because it is simple.
</details>

</details>

**Load Model**
``` python
from llava.model.builder import load_pretrained_model
from llava.constants import DEFAULT_IMAGE_TOKEN

from llava.mm_utils import (
    KeywordsStoppingCriteria,
    get_model_name_from_path,
    tokenizer_image_token,
    process_images
)
from llava.conversation import SeparatorStyle, conv_templates
from llava.eval.model_vqa import preprocess_qwen

overwrite_config = {}
overwrite_config["mm_spatial_pool_stride"] = 2
overwrite_config["mm_spatial_pool_mode"] = 'bilinear'
overwrite_config["mm_pooling_position"] = 'after'
overwrite_config["mm_newline_position"] = 'no_token'

model_path = "Inst-IT/LLaVA-Next-Inst-It-Qwen2-7B"
model_name = get_model_name_from_path(model_path)

tokenizer, model, image_processor, max_length = load_pretrained_model(
            model_path=model_path, 
            model_base=None, 
            model_name=model_name,
            device_map="auto", 
            torch_dtype='bfloat16', 
            overwrite_config=overwrite_config,
            attn_implementation='sdpa')
```
**Image Inference**

<details>
<summary>Inference without SoMs</summary>

Our model can perform inference on images without [Set-of-Marks](https://arxiv.org/abs/2310.11441) visual prompts, in this case, it can be used in the same way as its base mode [LLaVA-NeXT](https://github.com/LLaVA-VL/LLaVA-NeXT).

``` python
import torch
import requests
from PIL import Image

img_url = "https://github.com/inst-it/inst-it/blob/main/assets/demo/image.jpg?raw=true"
image = Image.open(requests.get(img_url, stream=True).raw)
image_tensor = process_images([image], image_processor, model.config).bfloat16()
image_sizes = [image.size]

question = "Describe this image."
question = DEFAULT_IMAGE_TOKEN + "\n" + question

conv_template = 'qwen_1_5'
conv = conv_templates[conv_template].copy()
conv.append_message(conv.roles[0], question)
conv.append_message(conv.roles[1], None)
prompt = conv.get_prompt()

input_ids = preprocess_qwen([{'from': 'human','value': question},{'from': 'gpt','value': None}], tokenizer, has_image=True).cuda()

pad_token_ids = tokenizer.pad_token_id if tokenizer.pad_token_id is not None else tokenizer.eos_token_id
attention_masks = input_ids.ne(pad_token_ids).long().cuda()

stop_str = conv.sep if conv.sep_style != SeparatorStyle.TWO else conv.sep2
keywords = [stop_str]
stopping_criteria = KeywordsStoppingCriteria(keywords, tokenizer, input_ids)

with torch.inference_mode():
    output_ids = model.generate(
        inputs=input_ids,
        images=image_tensor,
        attention_mask=attention_masks,
        modalities="image",
        image_sizes=image_sizes,
        use_cache=True,
        stopping_criteria=[stopping_criteria],
        max_new_tokens=4096
    )

pred = tokenizer.batch_decode(output_ids, skip_special_tokens=True)[0].strip()
print(pred)
```
</details>

<details>
<summary>Inference with SoMs</summary>

Our model performs more fine-grained understanding when [Set-of-Marks](https://arxiv.org/abs/2310.11441) visual prompts are provided. 
You can refer to the instances that you are interested in using their IDs.
Compared to the previous inference code, the following code has no modifications except for the input image, which is visual prompted with Set-of-Marks. 
Refer to [this link](https://github.com/microsoft/SoM) to learn how to generate SoMs for an image.

``` python
import torch
import requests
from PIL import Image

img_url = "https://github.com/inst-it/inst-it/blob/main/assets/demo/image_som.jpg?raw=true"
image = Image.open(requests.get(img_url, stream=True).raw)
image_tensor = process_images([image], image_processor, model.config).bfloat16()
image_sizes = [image.size]

# You can use [id] to refer to the instances that you are interested in
question = "Describe [8] in detail."
question = DEFAULT_IMAGE_TOKEN + "\n" + question

conv_template = 'qwen_1_5'
conv = conv_templates[conv_template].copy()
conv.append_message(conv.roles[0], question)
conv.append_message(conv.roles[1], None)
prompt = conv.get_prompt()

input_ids = preprocess_qwen([{'from': 'human','value': question},{'from': 'gpt','value': None}], tokenizer, has_image=True).cuda()

pad_token_ids = tokenizer.pad_token_id if tokenizer.pad_token_id is not None else tokenizer.eos_token_id
attention_masks = input_ids.ne(pad_token_ids).long().cuda()

stop_str = conv.sep if conv.sep_style != SeparatorStyle.TWO else conv.sep2
keywords = [stop_str]
stopping_criteria = KeywordsStoppingCriteria(keywords, tokenizer, input_ids)

with torch.inference_mode():
    output_ids = model.generate(
        inputs=input_ids,
        images=image_tensor,
        attention_mask=attention_masks,
        modalities="image",
        image_sizes=image_sizes,
        use_cache=True,
        stopping_criteria=[stopping_criteria],
        max_new_tokens=4096
    )

pred = tokenizer.batch_decode(output_ids, skip_special_tokens=True)[0].strip()
print(pred)
```
</details>

**Video Inference**

For the video, we organize each frame into a list. You can use the format \<t\> to refer to a specific timestamp (e.g. <1>).

<details>
<summary>Inference without SoMs</summary>

Our model can perform inference on videos without [Set-of-Marks](https://arxiv.org/abs/2310.11441) visual prompts, in this case, it can be used in the same way as its base mode [LLaVA-NeXT](https://github.com/LLaVA-VL/LLaVA-NeXT).

``` python
import torch
import requests
from PIL import Image

frame_urls = [
    "https://github.com/inst-it/inst-it/blob/main/assets/demo/frame_1.jpg?raw=true",
    "https://github.com/inst-it/inst-it/blob/main/assets/demo/frame_2.jpg?raw=true",
    "https://github.com/inst-it/inst-it/blob/main/assets/demo/frame_3.jpg?raw=true",
    "https://github.com/inst-it/inst-it/blob/main/assets/demo/frame_4.jpg?raw=true",
    "https://github.com/inst-it/inst-it/blob/main/assets/demo/frame_5.jpg?raw=true",
    "https://github.com/inst-it/inst-it/blob/main/assets/demo/frame_6.jpg?raw=true",
    "https://github.com/inst-it/inst-it/blob/main/assets/demo/frame_7.jpg?raw=true",
    "https://github.com/inst-it/inst-it/blob/main/assets/demo/frame_8.jpg?raw=true"
]
video = [Image.open(requests.get(frame_url, stream=True).raw) for frame_url in frame_urls]
video = image_processor.preprocess(video, return_tensors="pt")["pixel_values"].cuda()
video = video.bfloat16()
videos = [video]

question = "Describe the video."  # overall video caption
question = "What happens at frame <1>?"  # caption a specific moment
question = DEFAULT_IMAGE_TOKEN + "\n" + question

conv_template = 'qwen_1_5'
conv = conv_templates[conv_template].copy()
conv.append_message(conv.roles[0], question)
conv.append_message(conv.roles[1], None)
prompt = conv.get_prompt()

input_ids = preprocess_qwen([{'from': 'human','value': question},{'from': 'gpt','value': None}], tokenizer, has_image=True).cuda()

pad_token_ids = tokenizer.pad_token_id if tokenizer.pad_token_id is not None else tokenizer.eos_token_id
attention_masks = input_ids.ne(pad_token_ids).long().cuda()

stop_str = conv.sep if conv.sep_style != SeparatorStyle.TWO else conv.sep2
keywords = [stop_str]
stopping_criteria = KeywordsStoppingCriteria(keywords, tokenizer, input_ids)

with torch.inference_mode():
    output_ids = model.generate(
        inputs=input_ids,
        images=videos,
        attention_mask=attention_masks,
        modalities="video",
        use_cache=True,
        stopping_criteria=[stopping_criteria],
        max_new_tokens=4096
    )

pred = tokenizer.batch_decode(output_ids, skip_special_tokens=True)[0].strip()
print(pred)
```
</details>

<details>
<summary>Inference with SoMs</summary>

Our model performs more fine-grained understanding when [Set-of-Marks](https://arxiv.org/abs/2310.11441) visual prompts are provided. 
You can refer to the instances that you are interested in using their IDs.
Compared to the previous inference code, the following code has no modifications except for the input video, which is visual prompted with Set-of-Marks. 
Refer to [SAM2](https://github.com/facebookresearch/sam2) and [SoM](https://github.com/microsoft/SoM) to learn how to generate SoMs for a video.

``` python
import torch
import requests
from PIL import Image

frame_urls = [
    "https://github.com/inst-it/inst-it/blob/main/assets/demo/som_frame_1.jpg?raw=true",
    "https://github.com/inst-it/inst-it/blob/main/assets/demo/som_frame_2.jpg?raw=true",
    "https://github.com/inst-it/inst-it/blob/main/assets/demo/som_frame_3.jpg?raw=true",
    "https://github.com/inst-it/inst-it/blob/main/assets/demo/som_frame_4.jpg?raw=true",
    "https://github.com/inst-it/inst-it/blob/main/assets/demo/som_frame_5.jpg?raw=true",
    "https://github.com/inst-it/inst-it/blob/main/assets/demo/som_frame_6.jpg?raw=true",
    "https://github.com/inst-it/inst-it/blob/main/assets/demo/som_frame_7.jpg?raw=true",
    "https://github.com/inst-it/inst-it/blob/main/assets/demo/som_frame_8.jpg?raw=true"
]
video = [Image.open(requests.get(frame_url, stream=True).raw) for frame_url in frame_urls]
video = image_processor.preprocess(video, return_tensors="pt")["pixel_values"].cuda()
video = video.bfloat16()
videos = [video]

# You can use [id] to refer to the instances that you are interested in
question = "Is [3] visible at <1>?"
question = DEFAULT_IMAGE_TOKEN + "\n" + question

conv_template = 'qwen_1_5'
conv = conv_templates[conv_template].copy()
conv.append_message(conv.roles[0], question)
conv.append_message(conv.roles[1], None)
prompt = conv.get_prompt()

input_ids = preprocess_qwen([{'from': 'human','value': question},{'from': 'gpt','value': None}], tokenizer, has_image=True).cuda()

pad_token_ids = tokenizer.pad_token_id if tokenizer.pad_token_id is not None else tokenizer.eos_token_id
attention_masks = input_ids.ne(pad_token_ids).long().cuda()

stop_str = conv.sep if conv.sep_style != SeparatorStyle.TWO else conv.sep2
keywords = [stop_str]
stopping_criteria = KeywordsStoppingCriteria(keywords, tokenizer, input_ids)

with torch.inference_mode():
    output_ids = model.generate(
        inputs=input_ids,
        images=videos,
        attention_mask=attention_masks,
        modalities="video",
        use_cache=True,
        stopping_criteria=[stopping_criteria],
        max_new_tokens=4096
    )

pred = tokenizer.batch_decode(output_ids, skip_special_tokens=True)[0].strip()
print(pred)
```
</details>

## Contact
Feel free to contact us if you have any questions or suggestions 
- Email (Wujian Peng): [email protected]
- Email (Lingchen Meng): [email protected]

## Citation
``` bibtex
@article{peng2024inst,
  title={Inst-IT: Boosting Multimodal Instance Understanding via Explicit Visual Prompt Instruction Tuning},
  author={Peng, Wujian and Meng, Lingchen and Chen, Yitong and Xie, Yiweng and Liu, Yang and Gui, Tao and Xu, Hang and Qiu, Xipeng and Wu, Zuxuan and Jiang, Yu-Gang},
  journal={arXiv preprint arXiv:2412.03565},
  year={2024}
}
```